# EveryLLM > Every large language model in the world. A free guide to 471 models, filed by kind (general, reasoning, code, multimodal, small, domain, regional, encoder, research, assistant); access (open, community, open-source, api, research, retired); architecture (decoder, moe, encoder, encoder-decoder, ssm, hybrid, diffusion); modality (text, vision, audio, video, image-out, tools); language (multilingual, english, chinese, indic, arabic, european, japanese, korean, other); by parameters on a 0 to 5 scale (0 small, under 1 billion; 1 compact, 1 to 10 billion; 2 mid-size, 10 to 70 billion; 3 large, 70 to 300 billion; 4 very large, 300 billion to 1 trillion; 5 trillion-scale, 1 trillion and more); by developers, region, country and date. With hubs, a calendar, a timeline, a world map and a glossary. Published by ClearTrust. Each page gives: name, developed by, parameters, active parameters, context window, training data, training compute, design, licence, inputs and outputs, came after, followed by, what it is, how it came about, inside the model, what it can do, how to use it, safety and testing, reception and debate, why it matters, a short FAQ, further reading and source links. Pages are derived from Wikipedia and Wikidata with AI assistance and may contain mistakes; cite the page URL. ## Browse - [All models](https://everyllm.cpn.cleartrust.cc/models) - [Timeline](https://everyllm.cpn.cleartrust.cc/timeline) - [AI lab cities](https://everyllm.cpn.cleartrust.cc/hubs) - [Regions](https://everyllm.cpn.cleartrust.cc/regions) - [Countries](https://everyllm.cpn.cleartrust.cc/countries) - [World map](https://everyllm.cpn.cleartrust.cc/map) - [General-purpose models](https://everyllm.cpn.cleartrust.cc/kinds/general) - [Reasoning models](https://everyllm.cpn.cleartrust.cc/kinds/reasoning) - [Code models](https://everyllm.cpn.cleartrust.cc/kinds/code) - [Multimodal models](https://everyllm.cpn.cleartrust.cc/kinds/multimodal) - [Small and on-device](https://everyllm.cpn.cleartrust.cc/kinds/small) - [Domain models](https://everyllm.cpn.cleartrust.cc/kinds/domain) - [Regional and language models](https://everyllm.cpn.cleartrust.cc/kinds/regional) - [Encoders and embeddings](https://everyllm.cpn.cleartrust.cc/kinds/encoder) - [Research milestones](https://everyllm.cpn.cleartrust.cc/kinds/research) - [Assistants and chatbots](https://everyllm.cpn.cleartrust.cc/kinds/assistant) - [Open weights, permissive licence](https://everyllm.cpn.cleartrust.cc/access/open) - [Open weights with conditions](https://everyllm.cpn.cleartrust.cc/access/community) - [Fully open](https://everyllm.cpn.cleartrust.cc/access/open-source) - [API and app only](https://everyllm.cpn.cleartrust.cc/access/api) - [Research access only](https://everyllm.cpn.cleartrust.cc/access/research) - [Retired](https://everyllm.cpn.cleartrust.cc/access/retired) - [Dense decoder transformers](https://everyllm.cpn.cleartrust.cc/architectures/decoder) - [Mixture of experts](https://everyllm.cpn.cleartrust.cc/architectures/moe) - [Encoder-only](https://everyllm.cpn.cleartrust.cc/architectures/encoder) - [Encoder-decoder](https://everyllm.cpn.cleartrust.cc/architectures/encoder-decoder) - [State-space and recurrent](https://everyllm.cpn.cleartrust.cc/architectures/ssm) - [Hybrid designs](https://everyllm.cpn.cleartrust.cc/architectures/hybrid) - [Diffusion language models](https://everyllm.cpn.cleartrust.cc/architectures/diffusion) - [Text](https://everyllm.cpn.cleartrust.cc/modalities/text) - [Images in](https://everyllm.cpn.cleartrust.cc/modalities/vision) - [Speech and audio](https://everyllm.cpn.cleartrust.cc/modalities/audio) - [Video in](https://everyllm.cpn.cleartrust.cc/modalities/video) - [Images out](https://everyllm.cpn.cleartrust.cc/modalities/image-out) - [Tools and agents](https://everyllm.cpn.cleartrust.cc/modalities/tools) - [Many languages](https://everyllm.cpn.cleartrust.cc/languages/multilingual) - [English first](https://everyllm.cpn.cleartrust.cc/languages/english) - [Chinese](https://everyllm.cpn.cleartrust.cc/languages/chinese) - [Indian languages](https://everyllm.cpn.cleartrust.cc/languages/indic) - [Arabic](https://everyllm.cpn.cleartrust.cc/languages/arabic) - [European languages](https://everyllm.cpn.cleartrust.cc/languages/european) - [Japanese](https://everyllm.cpn.cleartrust.cc/languages/japanese) - [Korean](https://everyllm.cpn.cleartrust.cc/languages/korean) - [Other languages](https://everyllm.cpn.cleartrust.cc/languages/other) ## Live - [Today](https://everyllm.cpn.cleartrust.cc/today): on this day, model of the day - [Calendar](https://everyllm.cpn.cleartrust.cc/calendar) - [News](https://everyllm.cpn.cleartrust.cc/news) ## Learn - [Concepts](https://everyllm.cpn.cleartrust.cc/concepts) ## For agents - [MCP server](https://everyllm.cpn.cleartrust.cc/mcp): Streamable HTTP, no key needed - [JSON API](https://everyllm.cpn.cleartrust.cc/developers): GET https://everyllm.cpn.cleartrust.cc/api/v1/ - [Sitemap](https://everyllm.cpn.cleartrust.cc/sitemap.xml) - [How complete the catalogue is](https://everyllm.cpn.cleartrust.cc/coverage) ## AI lab cities - [San Francisco](https://everyllm.cpn.cleartrust.cc/hub/san-francisco): Home of OpenAI and Anthropic, and the densest cluster of frontier AI labs in the world. - [Mountain View and Palo Alto](https://everyllm.cpn.cleartrust.cc/hub/mountain-view-palo-alto): Google's headquarters and Stanford University: where the transformer, BERT and Gemini were born. - [Santa Clara](https://everyllm.cpn.cleartrust.cc/hub/santa-clara): Nvidia's home town, whose chips train most large models and whose researchers build Nemotron. - [Berkeley](https://everyllm.cpn.cleartrust.cc/hub/berkeley): UC Berkeley's labs gave the world Vicuna, Chatbot Arena and vLLM, a common engine for serving models. - [Menlo Park](https://everyllm.cpn.cleartrust.cc/hub/menlo-park): Meta's headquarters, home of the Llama open-weights models and their successor, Muse. - [Seattle and Redmond](https://everyllm.cpn.cleartrust.cc/hub/seattle-redmond): Microsoft, Amazon and the Allen Institute for AI: Copilot, Phi, Amazon Q and the fully open OLMo. - [New York](https://everyllm.cpn.cleartrust.cc/hub/new-york): Meta's FAIR lab, IBM, Hugging Face and the universities, and the city where AI copyright cases are fought. - [Boston and Cambridge](https://everyllm.cpn.cleartrust.cc/hub/boston-cambridge): MIT and Harvard, birthplace of ELIZA, now home to new model start-ups and large open datasets. - [Pittsburgh](https://everyllm.cpn.cleartrust.cc/hub/pittsburgh): Carnegie Mellon University: XLNet, Mamba and decades of work on speech and language. - [Toronto](https://everyllm.cpn.cleartrust.cc/hub/toronto): Geoffrey Hinton's city: the Vector Institute, Cohere and a deep line of neural-network research. - [Montreal](https://everyllm.cpn.cleartrust.cc/hub/montreal): Yoshua Bengio's Mila: neural language models, attention for translation and AI safety research. - [London](https://everyllm.cpn.cleartrust.cc/hub/london): Google DeepMind's home at King's Cross: Gopher, Chinchilla, Sparrow and much of Gemini. - [Oxford](https://everyllm.cpn.cleartrust.cc/hub/oxford): A university city strong in machine learning, AI governance and research on model collapse. - [Cambridge (UK)](https://everyllm.cpn.cleartrust.cc/hub/cambridge-uk): University and Microsoft Research labs, Arm's chip designs and Amazon's Alexa team. - [Paris](https://everyllm.cpn.cleartrust.cc/hub/paris): Mistral AI, Hugging Face, Kyutai and Meta's FAIR lab: Europe's capital of open models. - [Berlin](https://everyllm.cpn.cleartrust.cc/hub/berlin): Germany's start-up capital, with AI research institutes and the national policy debate. - [Munich](https://everyllm.cpn.cleartrust.cc/hub/munich): LMU, TUM and the Munich Center for Machine Learning, with big-tech research offices nearby. - [Heidelberg](https://everyllm.cpn.cleartrust.cc/hub/heidelberg): Home of Aleph Alpha, Germany's best-known model builder and a test of European AI sovereignty. - [Zurich](https://everyllm.cpn.cleartrust.cc/hub/zurich): ETH Zurich and the Swiss AI Initiative behind Apertus, plus research offices of Google and OpenAI. - [Lausanne](https://everyllm.cpn.cleartrust.cc/hub/lausanne): EPFL, co-developer of the Swiss Apertus model and a centre for open and medical AI research. - [Amsterdam](https://everyllm.cpn.cleartrust.cc/hub/amsterdam): The Netherlands' AI hub, where the public GPT-NL model is trained on the national supercomputer. - [Helsinki](https://everyllm.cpn.cleartrust.cc/hub/helsinki): Silo AI and the University of Turku built the Poro and Viking models for Nordic languages on LUMI. - [Stockholm](https://everyllm.cpn.cleartrust.cc/hub/stockholm): Sweden's technology capital, close to the AI Sweden effort behind GPT-SW3 for the Nordic languages. - [Rome](https://everyllm.cpn.cleartrust.cc/hub/rome): Sapienza University's Minerva and Almawave's Velvet: Italy's home-grown Italian-language models. - [Lisbon](https://everyllm.cpn.cleartrust.cc/hub/lisbon): Portugal's Amália model and the Lisbon-led EuroLLM project for all EU languages. - [Moscow](https://everyllm.cpn.cleartrust.cc/hub/moscow): Sber's GigaChat and Yandex's Alice AI: Russia's own chatbots and model families. - [Tel Aviv](https://everyllm.cpn.cleartrust.cc/hub/tel-aviv): AI21 Labs, maker of the Jurassic and Jamba models, and research offices of the large US companies. - [Abu Dhabi](https://everyllm.cpn.cleartrust.cc/hub/abu-dhabi): TII's Falcon, the Arabic Jais, MBZUAI's K2 Think and G42: a state-backed bid for top models. - [Riyadh](https://everyllm.cpn.cleartrust.cc/hub/riyadh): SDAIA's ALLaM and the national AI company Humain: Saudi Arabia's Arabic-first model programme. - [Bengaluru](https://everyllm.cpn.cleartrust.cc/hub/bengaluru): Sarvam AI, Ola Krutrim and the research labs of IISc, Google and Microsoft: India's model capital. - [Mumbai](https://everyllm.cpn.cleartrust.cc/hub/mumbai): IIT Bombay leads BharatGen, India's largest government-funded model consortium; Yotta's GPUs sit nearby. - [Chennai](https://everyllm.cpn.cleartrust.cc/hub/chennai): IIT Madras and AI4Bharat build open data, translation and speech models for all 22 scheduled languages. - [Hyderabad](https://everyllm.cpn.cleartrust.cc/hub/hyderabad): IIIT Hyderabad and IIT Hyderabad research Indian-language AI; Microsoft, Google and Amazon run big centres. - [Delhi NCR](https://everyllm.cpn.cleartrust.cc/hub/delhi-ncr): Seat of the IndiaAI Mission and host of the India AI Impact Summit in February 2026. - [Beijing](https://everyllm.cpn.cleartrust.cc/hub/beijing): Zhipu, Moonshot AI, Baidu, ByteDance and BAAI: China's largest cluster of model builders. - [Hangzhou](https://everyllm.cpn.cleartrust.cc/hub/hangzhou): Alibaba's Qwen and DeepSeek: the city behind two of the world's most used open-weights families. - [Shanghai](https://everyllm.cpn.cleartrust.cc/hub/shanghai): MiniMax, StepFun, SenseTime and the Shanghai AI Laboratory, plus the yearly WAIC conference. - [Shenzhen](https://everyllm.cpn.cleartrust.cc/hub/shenzhen): Huawei's PanGu and Ascend chips and Tencent's Hunyuan: China's hardware and platform giants. - [Hong Kong](https://everyllm.cpn.cleartrust.cc/hub/hong-kong): Universities, SenseTime's headquarters and HKChat, a chatbot built for Hong Kong users. - [Seoul](https://everyllm.cpn.cleartrust.cc/hub/seoul): Naver's HyperCLOVA X, LG's EXAONE and Korea's national foundation model teams. - [Tokyo](https://everyllm.cpn.cleartrust.cc/hub/tokyo): Sakana AI, Preferred Networks, NTT, SoftBank and national labs building Japanese-language models. - [Singapore](https://everyllm.cpn.cleartrust.cc/hub/singapore): AI Singapore's SEA-LION and A*STAR's MERaLiON: models for Southeast Asia's many languages. - [Sydney](https://everyllm.cpn.cleartrust.cc/hub/sydney): Australia's largest city: CSIRO's Data61, universities and companies building on language models. - [Johannesburg](https://everyllm.cpn.cleartrust.cc/hub/johannesburg): Lelapa AI's InkubaLM and the Masakhane community: open models for Africa's languages. ## Calendar - [NeurIPS](https://everyllm.cpn.cleartrust.cc/events/neurips): The largest machine learning research conference, held each December. The 2026 edition is centred on Sydney. - [ICLR](https://everyllm.cpn.cleartrust.cc/events/iclr): The International Conference on Learning Representations, a spring meeting known for its open peer review. - [ICML](https://everyllm.cpn.cleartrust.cc/events/icml): The International Conference on Machine Learning, one of the field's oldest meetings, comes to Rio de Janeiro in 2027. - [AAAI Conference on Artificial Intelligence](https://everyllm.cpn.cleartrust.cc/events/aaai): The 41st AAAI conference, one of the broadest artificial intelligence meetings, at the Palais des congrès de Montréal. - [ACL Annual Meeting](https://everyllm.cpn.cleartrust.cc/events/acl): The yearly meeting of the Association for Computational Linguistics, the main conference on language technology. - [EMNLP](https://everyllm.cpn.cleartrust.cc/events/emnlp): Empirical Methods in Natural Language Processing, a leading language technology conference held each autumn. - [NAACL](https://everyllm.cpn.cleartrust.cc/events/naacl): The conference of the Nations of the Americas Chapter of the ACL, where ELMo and BERT won best paper awards. - [COLM](https://everyllm.cpn.cleartrust.cc/events/colm): The Conference on Language Modeling, the first academic conference devoted entirely to language models. - [CVPR](https://everyllm.cpn.cleartrust.cc/events/cvpr): The Conference on Computer Vision and Pattern Recognition, where much vision-language model research appears. - [COLING](https://everyllm.cpn.cleartrust.cc/events/coling): The International Conference on Computational Linguistics, held since 1965. The 32nd edition is in Macau. - [Google I/O](https://everyllm.cpn.cleartrust.cc/events/google-io): Google's yearly developer conference at Shoreline Amphitheatre, where it presents new Gemini models and features. - [Microsoft Build](https://everyllm.cpn.cleartrust.cc/events/microsoft-build): Microsoft's developer conference, focused in recent years on Copilot, agents and the models behind them. - [Apple WWDC](https://everyllm.cpn.cleartrust.cc/events/wwdc): Apple's Worldwide Developers Conference, a June week of software news that introduced Apple Intelligence in 2024. - [OpenAI DevDay](https://everyllm.cpn.cleartrust.cc/events/openai-devday): OpenAI's developer conference in San Francisco, where it announces new models and tools for its API. - [Meta Connect](https://everyllm.cpn.cleartrust.cc/events/meta-connect): Meta's yearly event for glasses, headsets and AI, where it released the Llama 3.2 models in 2024. - [Nvidia GTC](https://everyllm.cpn.cleartrust.cc/events/nvidia-gtc): Nvidia's GPU Technology Conference in San Jose, where the chips that train language models are unveiled. - [AWS re:Invent](https://everyllm.cpn.cleartrust.cc/events/aws-reinvent): Amazon Web Services' yearly cloud conference in Las Vegas, where Amazon presents its models and AI services. - [PyTorch Conference](https://everyllm.cpn.cleartrust.cc/events/pytorch-conference): The yearly gathering of the open-source PyTorch community, the software most open models are built with. - [World Artificial Intelligence Conference](https://everyllm.cpn.cleartrust.cc/events/waic): Shanghai's yearly AI conference and exhibition, China's main stage for new models and AI governance. - [Bengaluru Tech Summit](https://everyllm.cpn.cleartrust.cc/events/bengaluru-tech-summit): Karnataka's yearly technology summit in Bengaluru, held in 2026 with the theme "AI & Beyond". - [AI Summit, Geneva 2027](https://everyllm.cpn.cleartrust.cc/events/geneva-ai-summit): The next government-led global AI summit, after Bletchley Park, Seoul, Paris and New Delhi, hosted by Switzerland. - [India AI Impact Summit anniversary](https://everyllm.cpn.cleartrust.cc/events/india-ai-impact-summit): On 16 February 2026 the India AI Impact Summit opened at Bharat Mandapam, the first of the series in the Global South. - [AI Safety Summit anniversary](https://everyllm.cpn.cleartrust.cc/events/bletchley-park-summit): On 1 November 2023 the first global AI Safety Summit opened at Bletchley Park in England. - ["Attention Is All You Need" published](https://everyllm.cpn.cleartrust.cc/events/attention-is-all-you-need): On 12 June 2017 eight Google researchers posted the paper that introduced the transformer. - [GPT-1 release](https://everyllm.cpn.cleartrust.cc/events/gpt-1-release): On 11 June 2018 OpenAI described GPT-1, the first generative pre-trained transformer. - [BERT published](https://everyllm.cpn.cleartrust.cc/events/bert-release): On 11 October 2018 Google researchers posted the BERT paper, which reset the records of language understanding. - [GPT-2 announced](https://everyllm.cpn.cleartrust.cc/events/gpt-2-release): On 14 February 2019 OpenAI announced GPT-2 and, citing misuse concerns, at first withheld the full model. - [GPT-3 paper published](https://everyllm.cpn.cleartrust.cc/events/gpt-3-paper): On 28 May 2020 OpenAI posted "Language Models are Few-Shot Learners", describing the 175-billion-parameter GPT-3. - [ChatGPT launch](https://everyllm.cpn.cleartrust.cc/events/chatgpt-launch): On 30 November 2022 OpenAI released ChatGPT as a free research preview, bringing language models to the public. - [LLaMA release](https://everyllm.cpn.cleartrust.cc/events/llama-release): On 24 February 2023 Meta released LLaMA to researchers. Its weights spread online within a week. - [GPT-4 release](https://everyllm.cpn.cleartrust.cc/events/gpt-4-release): On 14 March 2023 OpenAI released GPT-4, which accepted images as well as text. - [Claude release](https://everyllm.cpn.cleartrust.cc/events/claude-release): On 14 March 2023, the same day as GPT-4, Anthropic made its first Claude models available. - [Llama 2 release](https://everyllm.cpn.cleartrust.cc/events/llama-2-release): On 18 July 2023 Meta released Llama 2 with weights free for most commercial use. - [Gemini launch](https://everyllm.cpn.cleartrust.cc/events/gemini-launch): On 6 December 2023 Google DeepMind announced Gemini 1.0, trained from the start on text, images, audio and video. - [Bard announced](https://everyllm.cpn.cleartrust.cc/events/bard-announced): On 6 February 2023 Google announced Bard, its answer to ChatGPT, later renamed Gemini. - [Gemma release](https://everyllm.cpn.cleartrust.cc/events/gemma-release): On 21 February 2024 Google released Gemma, open-weight models built from its Gemini research. - [GPT-4o announced](https://everyllm.cpn.cleartrust.cc/events/gpt-4o-release): On 13 May 2024 OpenAI announced GPT-4o, one model for text, images and speech, and opened it to free users. - [OpenAI o1 preview](https://everyllm.cpn.cleartrust.cc/events/openai-o1-release): On 12 September 2024 OpenAI previewed o1, the first widely available reasoning model. - [DeepSeek-V3 release](https://everyllm.cpn.cleartrust.cc/events/deepseek-v3-release): On 26 December 2024 DeepSeek released V3, a 671-billion-parameter open mixture-of-experts model. - [DeepSeek-R1 release](https://everyllm.cpn.cleartrust.cc/events/deepseek-r1-release): On 20 January 2025 DeepSeek released R1, an open reasoning model under the MIT licence. - [GPT-5 release](https://everyllm.cpn.cleartrust.cc/events/gpt-5-release): On 7 August 2025 OpenAI released GPT-5, which switches between quick answers and longer reasoning. ## Concepts - **Large language model**: A neural network trained on very large amounts of text to predict the next token, and so to read and write language. (https://everyllm.cpn.cleartrust.cc/concepts/large-language-model) - **Language model**: Any system that assigns probabilities to sequences of words, and so can predict or generate text. (https://everyllm.cpn.cleartrust.cc/concepts/language-model) - **Token**: The unit of text a model reads and writes: a whole word, a piece of a word, a punctuation mark or a byte. (https://everyllm.cpn.cleartrust.cc/concepts/token) - **Tokenizer**: The program that splits text into tokens and turns them into numbers before a model sees them, and back again afterwards. (https://everyllm.cpn.cleartrust.cc/concepts/tokenizer) - **Byte-pair encoding**: A method for building a vocabulary by repeatedly merging the most frequent pair of symbols in a body of text. (https://everyllm.cpn.cleartrust.cc/concepts/byte-pair-encoding) - **SentencePiece**: A tokenizer library from Google that treats text as a raw stream, spaces included, so it works for languages written without spaces. (https://everyllm.cpn.cleartrust.cc/concepts/sentencepiece) - **Vocabulary**: The fixed list of tokens a model knows, each with its own number and its own embedding. (https://everyllm.cpn.cleartrust.cc/concepts/vocabulary) - **Special token**: A reserved token that marks structure rather than text, such as the end of a document or the start of a user's turn. (https://everyllm.cpn.cleartrust.cc/concepts/special-token) - **Tokenizer fertility**: How many tokens a tokenizer needs, on average, for one word of a given language. Lower is cheaper and faster. (https://everyllm.cpn.cleartrust.cc/concepts/tokenizer-fertility) - **Embedding**: A list of numbers that stands for a token, word or passage, placed so that similar meanings sit close together. (https://everyllm.cpn.cleartrust.cc/concepts/embedding) - **Word2vec**: A 2013 Google method for learning word embeddings from which much of modern language modelling grew. (https://everyllm.cpn.cleartrust.cc/concepts/word2vec) - **ELMo**: A 2018 model from the Allen Institute for AI that produced contextual word embeddings from a two-way recurrent network. (https://everyllm.cpn.cleartrust.cc/concepts/elmo) - **Neural network**: A computing system made of layers of simple units whose connection strengths are learned from data. (https://everyllm.cpn.cleartrust.cc/concepts/neural-network) - **Transformer**: The neural network design, introduced in 2017, that almost every modern language model is built on. (https://everyllm.cpn.cleartrust.cc/concepts/transformer) - **Attention**: The step in a transformer where each token weighs every other token and pulls in information from the ones that matter. (https://everyllm.cpn.cleartrust.cc/concepts/attention) - **Self-attention**: Attention in which the tokens of one text attend to each other, rather than to a separate input. (https://everyllm.cpn.cleartrust.cc/concepts/self-attention) - **Multi-head attention**: Running several attention operations side by side, so different heads can track different kinds of relationship. (https://everyllm.cpn.cleartrust.cc/concepts/multi-head-attention) - **Grouped-query attention**: A form of attention in which several query heads share one set of keys and values, shrinking the KV cache. (https://everyllm.cpn.cleartrust.cc/concepts/grouped-query-attention) - **Multi-head latent attention**: DeepSeek's attention design that compresses keys and values into a small shared vector to save memory. (https://everyllm.cpn.cleartrust.cc/concepts/multi-head-latent-attention) - **FlashAttention**: A way of computing exact attention that avoids slow trips to GPU memory, making long contexts faster and cheaper. (https://everyllm.cpn.cleartrust.cc/concepts/flash-attention) - **Sliding-window attention**: Attention limited to a fixed number of recent tokens, so cost stops growing with the full length of the text. (https://everyllm.cpn.cleartrust.cc/concepts/sliding-window-attention) - **Positional encoding**: Information added to token embeddings so the model knows the order of words, which attention alone ignores. (https://everyllm.cpn.cleartrust.cc/concepts/positional-encoding) - **Rotary position embedding**: A way of encoding word order by rotating query and key vectors by an angle that depends on position. (https://everyllm.cpn.cleartrust.cc/concepts/rope) - **Layer**: One repeated block of a model. Text passes through dozens of layers, each refining the representation a little. (https://everyllm.cpn.cleartrust.cc/concepts/layer) - **Hidden state**: The vector a model holds for each token between layers, carrying everything it has worked out so far. (https://everyllm.cpn.cleartrust.cc/concepts/hidden-state) - **Feed-forward network**: The part of each transformer layer that processes every token on its own, holding much of the model's stored knowledge. (https://everyllm.cpn.cleartrust.cc/concepts/feed-forward-network) - **Residual connection**: A shortcut that adds a layer's input to its output, so signals and gradients can pass through very deep networks. (https://everyllm.cpn.cleartrust.cc/concepts/residual-connection) - **Layer normalisation**: A step that rescales each token's vector to a steady size inside every layer, keeping training stable. (https://everyllm.cpn.cleartrust.cc/concepts/layer-normalisation) - **Softmax**: A function that turns a list of raw scores into probabilities that are all positive and add up to one. (https://everyllm.cpn.cleartrust.cc/concepts/softmax) - **Logits**: The raw scores a model gives every token in its vocabulary before they are turned into probabilities. (https://everyllm.cpn.cleartrust.cc/concepts/logits) - **Encoder**: A model, or part of one, that reads a whole text at once in both directions to build a representation of its meaning. (https://everyllm.cpn.cleartrust.cc/concepts/encoder) - **Decoder**: A model, or part of one, that writes text one token at a time, each token seeing only those before it. (https://everyllm.cpn.cleartrust.cc/concepts/decoder) - **Encoder-decoder**: A design with two stacks: one reads the input, the other writes the output while looking back at it. (https://everyllm.cpn.cleartrust.cc/concepts/encoder-decoder) - **Autoregressive generation**: Producing text one token at a time, feeding each new token back in as input for the next. (https://everyllm.cpn.cleartrust.cc/concepts/autoregressive) - **Next-token prediction**: The training task of guessing the next token of a text from all the tokens before it. (https://everyllm.cpn.cleartrust.cc/concepts/next-token-prediction) - **Masked language model**: A model trained to fill in hidden words in a sentence using the words on both sides, as BERT was. (https://everyllm.cpn.cleartrust.cc/concepts/masked-language-model) - **State-space model**: A family of sequence models that carry a fixed-size running state instead of attending to every past token. (https://everyllm.cpn.cleartrust.cc/concepts/state-space-model) - **Hybrid model**: A model that mixes attention layers with state-space or recurrent layers to get the strengths of both. (https://everyllm.cpn.cleartrust.cc/concepts/hybrid-model) - **Recurrent neural network**: A network that reads a sequence one step at a time, carrying a memory forward. The main language approach before transformers. (https://everyllm.cpn.cleartrust.cc/concepts/recurrent-neural-network) - **Diffusion language model**: A model that writes by starting from noise or masked text and refining the whole passage over several steps. (https://everyllm.cpn.cleartrust.cc/concepts/diffusion-language-model) - **Mixture of experts**: A design in which each layer holds many expert sub-networks and a router sends each token to just a few of them. (https://everyllm.cpn.cleartrust.cc/concepts/mixture-of-experts) - **Active parameters**: The parameters a mixture-of-experts model actually uses for each token, usually a small share of its total. (https://everyllm.cpn.cleartrust.cc/concepts/active-parameters) - **Parameter**: One adjustable number inside a model, learned in training. Model size is usually given as a parameter count. (https://everyllm.cpn.cleartrust.cc/concepts/parameter) - **Weights**: The full set of learned parameters of a model, saved as files. Having the weights means being able to run the model. (https://everyllm.cpn.cleartrust.cc/concepts/weights) - **Pre-training**: The first and largest stage of training, in which a model learns from trillions of tokens of general text. (https://everyllm.cpn.cleartrust.cc/concepts/pre-training) - **Base model**: A model straight out of pre-training, which continues text but has not yet been tuned to follow instructions. (https://everyllm.cpn.cleartrust.cc/concepts/base-model) - **Post-training**: Everything done to a model after pre-training: instruction tuning, preference tuning, safety training and reasoning training. (https://everyllm.cpn.cleartrust.cc/concepts/post-training) - **Fine-tuning**: Further training of an already trained model on a smaller, focused dataset to adapt it to a task or style. (https://everyllm.cpn.cleartrust.cc/concepts/fine-tuning) - **Supervised fine-tuning**: Fine-tuning on example prompts paired with good answers, usually the first step of turning a base model into an assistant. (https://everyllm.cpn.cleartrust.cc/concepts/supervised-fine-tuning) - **Instruction tuning**: Fine-tuning a model on many tasks phrased as instructions, so it learns to do what it is asked. (https://everyllm.cpn.cleartrust.cc/concepts/instruction-tuning) - **Reinforcement learning from human feedback**: Training a model with rewards learned from people's comparisons of its answers, the method that made ChatGPT helpful. (https://everyllm.cpn.cleartrust.cc/concepts/rlhf) - **Reward model**: A model trained to score answers the way human raters would, used to guide reinforcement learning. (https://everyllm.cpn.cleartrust.cc/concepts/reward-model) - **Direct preference optimisation**: A simpler alternative to RLHF that trains directly on preference pairs, with no separate reward model. (https://everyllm.cpn.cleartrust.cc/concepts/dpo) - **Constitutional AI**: Anthropic's method of training a model to critique and improve its own answers against a written list of principles. (https://everyllm.cpn.cleartrust.cc/concepts/constitutional-ai) - **Reinforcement learning**: Learning by trial and reward rather than from fixed examples. Used to tune chat models and to train reasoning models. (https://everyllm.cpn.cleartrust.cc/concepts/reinforcement-learning) - **Group relative policy optimisation**: A reinforcement learning method from DeepSeek that scores each answer against a group of answers to the same question. (https://everyllm.cpn.cleartrust.cc/concepts/grpo) - **Training data**: The text, code and other material a model learns from, now measured in trillions of tokens. (https://everyllm.cpn.cleartrust.cc/concepts/training-data) - **Data filtering**: Selecting and cleaning training text, removing spam, toxic content, personal data and low-quality pages. (https://everyllm.cpn.cleartrust.cc/concepts/data-filtering) - **Synthetic data**: Training examples written by other models rather than collected from people, now a large share of post-training data. (https://everyllm.cpn.cleartrust.cc/concepts/synthetic-data) - **Model collapse**: The loss of quality and variety when models are trained, generation after generation, on text written by earlier models. (https://everyllm.cpn.cleartrust.cc/concepts/model-collapse) - **Data contamination**: When benchmark questions or answers end up in a model's training data, inflating its scores. (https://everyllm.cpn.cleartrust.cc/concepts/data-contamination) - **Training compute**: The total amount of computation used to train a model, usually counted in floating-point operations. (https://everyllm.cpn.cleartrust.cc/concepts/training-compute) - **Graphics processing unit**: The chip that trains and runs language models, built for doing many simple calculations at once. (https://everyllm.cpn.cleartrust.cc/concepts/gpu) - **Mixed precision**: Training with lower-precision numbers for most calculations to save memory and time, while keeping key values precise. (https://everyllm.cpn.cleartrust.cc/concepts/mixed-precision) - **Cross-entropy loss**: The standard training loss for language models: how surprised the model was by the token that actually came next. (https://everyllm.cpn.cleartrust.cc/concepts/cross-entropy-loss) - **Gradient descent**: The basic training rule: move every parameter a small step in the direction that lowers the loss. (https://everyllm.cpn.cleartrust.cc/concepts/gradient-descent) - **Scaling laws**: Regular patterns showing that loss falls predictably as models, data and compute grow. (https://everyllm.cpn.cleartrust.cc/concepts/scaling-laws) - **Chinchilla scaling**: DeepMind's 2022 finding that, for a fixed budget, models should be trained on about 20 tokens per parameter. (https://everyllm.cpn.cleartrust.cc/concepts/chinchilla-scaling) - **Emergent abilities**: Skills that seem to appear suddenly once a model passes a certain size, and the debate over whether they are real. (https://everyllm.cpn.cleartrust.cc/concepts/emergent-abilities) - **Foundation model**: A large model trained on broad data that can be adapted to many tasks. A wider term than language model. (https://everyllm.cpn.cleartrust.cc/concepts/foundation-model) - **Frontier model**: One of the most capable models at a given time, usually from a handful of well-funded labs. (https://everyllm.cpn.cleartrust.cc/concepts/frontier-model) - **Small language model**: A compact model, usually under about 10 billion parameters, that can run on a laptop, a phone or modest servers. (https://everyllm.cpn.cleartrust.cc/concepts/small-language-model) - **On-device AI**: Running a model on the user's own phone, laptop or car rather than in a data centre. (https://everyllm.cpn.cleartrust.cc/concepts/on-device-ai) - **Knowledge distillation**: Training a smaller student model to imitate a larger teacher model's outputs. (https://everyllm.cpn.cleartrust.cc/concepts/distillation) - **Quantisation**: Storing a model's weights with fewer bits, such as 8 or 4 instead of 16, to cut memory and speed up inference. (https://everyllm.cpn.cleartrust.cc/concepts/quantisation) - **Low-rank adaptation**: A cheap fine-tuning method that freezes the model and trains small add-on matrices instead. (https://everyllm.cpn.cleartrust.cc/concepts/lora) - **QLoRA**: LoRA fine-tuning on top of a 4-bit quantised model, which let large models be tuned on a single GPU. (https://everyllm.cpn.cleartrust.cc/concepts/qlora) - **Inference**: Running a trained model to get answers, as opposed to training it. Most of the world's AI computing is now inference. (https://everyllm.cpn.cleartrust.cc/concepts/inference) - **KV cache**: Stored keys and values from earlier tokens, kept so the model does not recompute them for every new token. (https://everyllm.cpn.cleartrust.cc/concepts/kv-cache) - **Continuous batching**: Serving many users at once by adding and removing requests from a running batch at every generation step. (https://everyllm.cpn.cleartrust.cc/concepts/continuous-batching) - **Speculative decoding**: Speeding up generation by letting a small fast model draft several tokens and the big model check them in one pass. (https://everyllm.cpn.cleartrust.cc/concepts/speculative-decoding) - **Latency**: How long a model takes to respond: the wait for the first token and the speed of the tokens after it. (https://everyllm.cpn.cleartrust.cc/concepts/latency) - **Sampling**: Choosing each next token from the model's probabilities, with some randomness, rather than always taking the top one. (https://everyllm.cpn.cleartrust.cc/concepts/sampling) - **Temperature**: A setting that controls how adventurous sampling is. Low temperature gives safe, predictable text; high gives varied text. (https://everyllm.cpn.cleartrust.cc/concepts/temperature) - **Top-p sampling**: Sampling only from the smallest group of tokens whose probabilities add up to a set share, such as 90 per cent. (https://everyllm.cpn.cleartrust.cc/concepts/top-p) - **Greedy decoding**: Always picking the single most likely next token. Fast and repeatable, but prone to dull or looping text. (https://everyllm.cpn.cleartrust.cc/concepts/greedy-decoding) - **Perplexity**: A measure of how well a model predicts a text. Lower means the model was less surprised. (https://everyllm.cpn.cleartrust.cc/concepts/perplexity) - **Context window**: The maximum number of tokens a model can take in at once, counting the prompt, any documents and its own reply. (https://everyllm.cpn.cleartrust.cc/concepts/context-window) - **Long context**: The ability to read and use very long inputs, from hundreds of thousands to millions of tokens. (https://everyllm.cpn.cleartrust.cc/concepts/long-context) - **Needle in a haystack test**: A long-context test that hides one fact in a long text and asks the model to find it. (https://everyllm.cpn.cleartrust.cc/concepts/needle-in-a-haystack) - **Lost in the middle**: The tendency of models to use information at the start and end of a long input better than information in the middle. (https://everyllm.cpn.cleartrust.cc/concepts/lost-in-the-middle) - **Prompt**: The text, and any images or files, given to a model to respond to. (https://everyllm.cpn.cleartrust.cc/concepts/prompt) - **Prompt engineering**: Writing and testing prompts to get better, more reliable answers from a model. (https://everyllm.cpn.cleartrust.cc/concepts/prompt-engineering) - **System prompt**: Hidden instructions placed before the conversation that set a model's role, rules and tone. (https://everyllm.cpn.cleartrust.cc/concepts/system-prompt) - **In-context learning**: A model picking up a new task from examples or instructions in the prompt, without any change to its weights. (https://everyllm.cpn.cleartrust.cc/concepts/in-context-learning) - **Few-shot prompting**: Putting a handful of worked examples in the prompt so the model copies the pattern. (https://everyllm.cpn.cleartrust.cc/concepts/few-shot) - **Zero-shot**: Asking a model to do a task with no examples, just an instruction. (https://everyllm.cpn.cleartrust.cc/concepts/zero-shot) - **Prompt caching**: Saving the processed form of a long, repeated prompt so later requests that start the same way are cheaper and faster. (https://everyllm.cpn.cleartrust.cc/concepts/prompt-caching) - **Knowledge cutoff**: The date after which a model has seen no training data, so it knows nothing of later events unless told. (https://everyllm.cpn.cleartrust.cc/concepts/knowledge-cutoff) - **Chain of thought**: Having a model write out intermediate reasoning steps before its answer, which improves accuracy on hard problems. (https://everyllm.cpn.cleartrust.cc/concepts/chain-of-thought) - **Reasoning model**: A model trained to think at length, in a hidden or visible chain of thought, before giving its final answer. (https://everyllm.cpn.cleartrust.cc/concepts/reasoning-model) - **Test-time compute**: Spending more computation while answering, for example by thinking longer, to get better results. (https://everyllm.cpn.cleartrust.cc/concepts/test-time-compute) - **Self-consistency**: Asking a model the same question several times and taking the most common answer. (https://everyllm.cpn.cleartrust.cc/concepts/self-consistency) - **Process reward model**: A model that scores each step of a solution, not just the final answer, to guide reasoning. (https://everyllm.cpn.cleartrust.cc/concepts/process-reward-model) - **Hallucination**: When a model states something false or made up, such as a fake citation, with apparent confidence. (https://everyllm.cpn.cleartrust.cc/concepts/hallucination) - **Grounding**: Tying a model's answer to specific sources, such as documents or search results, rather than its memory alone. (https://everyllm.cpn.cleartrust.cc/concepts/grounding) - **Retrieval-augmented generation**: Searching a collection of documents for relevant passages and giving them to the model along with the question. (https://everyllm.cpn.cleartrust.cc/concepts/retrieval-augmented-generation) - **Embedding model**: A model that turns a passage of text into a single vector, used for search, clustering and retrieval. (https://everyllm.cpn.cleartrust.cc/concepts/embedding-model) - **Vector database**: A database built to store embeddings and quickly find those nearest to a query vector. (https://everyllm.cpn.cleartrust.cc/concepts/vector-database) - **Tool use**: A model asking an outside program, such as a calculator, search engine or database, to do part of a task. (https://everyllm.cpn.cleartrust.cc/concepts/tool-use) - **Function calling**: An API feature in which a model returns a structured request to run a named function with given arguments. (https://everyllm.cpn.cleartrust.cc/concepts/function-calling) - **Structured output**: Making a model reply in an exact machine-readable format, such as JSON matching a given schema. (https://everyllm.cpn.cleartrust.cc/concepts/structured-output) - **Agent**: A model that works towards a goal over many steps, choosing and using tools and reacting to the results. (https://everyllm.cpn.cleartrust.cc/concepts/agent) - **Computer use**: A model operating a computer as a person would, looking at the screen and moving the mouse and keyboard. (https://everyllm.cpn.cleartrust.cc/concepts/computer-use) - **Coding agent**: An agent that reads a codebase, edits files, runs commands and tests, and iterates until a change works. (https://everyllm.cpn.cleartrust.cc/concepts/coding-agent) - **Model Context Protocol**: An open standard from Anthropic for connecting models and agents to tools and data sources. (https://everyllm.cpn.cleartrust.cc/concepts/model-context-protocol) - **Multimodal model**: A model that takes in, and sometimes produces, more than one kind of data: text, images, audio or video. (https://everyllm.cpn.cleartrust.cc/concepts/multimodal) - **Vision-language model**: A model that reads images together with text, answering questions about photos, documents and charts. (https://everyllm.cpn.cleartrust.cc/concepts/vision-language-model) - **Vision encoder**: The part of a multimodal model that turns an image into vectors, usually by cutting it into small patches. (https://everyllm.cpn.cleartrust.cc/concepts/vision-encoder) - **Speech model**: A model that turns speech into text, text into speech, or holds a spoken conversation directly. (https://everyllm.cpn.cleartrust.cc/concepts/speech-model) - **Evaluation**: Testing what a model can do and how it behaves, with benchmarks, human ratings and targeted test sets. (https://everyllm.cpn.cleartrust.cc/concepts/evaluation) - **Benchmark**: A fixed set of test questions with a scoring method, used to compare models on one kind of skill. (https://everyllm.cpn.cleartrust.cc/concepts/benchmark) - **MMLU**: A widely used 2020 benchmark of about 16,000 multiple-choice questions across 57 subjects. (https://everyllm.cpn.cleartrust.cc/concepts/mmlu) - **GPQA**: A 2023 benchmark of very hard science questions written by PhD experts so that web searching does not help. (https://everyllm.cpn.cleartrust.cc/concepts/gpqa) - **HumanEval**: A 2021 coding benchmark of 164 Python problems, each checked by hidden unit tests. (https://everyllm.cpn.cleartrust.cc/concepts/humaneval) - **SWE-bench**: A benchmark of real bug reports from open-source Python projects, where a model must produce a fix that passes the tests. (https://everyllm.cpn.cleartrust.cc/concepts/swe-bench) - **Chatbot Arena**: A public site where people compare two anonymous models' answers, used to rank chat models by preference. (https://everyllm.cpn.cleartrust.cc/concepts/chatbot-arena) - **Benchmark saturation**: When top models score so close to the maximum on a benchmark that it can no longer tell them apart. (https://everyllm.cpn.cleartrust.cc/concepts/benchmark-saturation) - **Calibration**: Whether a model's confidence matches how often it is right. (https://everyllm.cpn.cleartrust.cc/concepts/calibration) - **Alignment**: Making AI systems pursue the goals and values their designers and users intend, and avoid harmful behaviour. (https://everyllm.cpn.cleartrust.cc/concepts/alignment) - **AI safety**: The work of preventing harm from AI systems, from everyday misuse to severe risks from very capable models. (https://everyllm.cpn.cleartrust.cc/concepts/ai-safety) - **Red-teaming**: Deliberately attacking a model to find ways it can be made to misbehave, before others find them. (https://everyllm.cpn.cleartrust.cc/concepts/red-teaming) - **Jailbreak**: A prompt crafted to get around a model's safety training and make it produce what it would normally refuse. (https://everyllm.cpn.cleartrust.cc/concepts/jailbreak) - **Prompt injection**: Hiding instructions in content a model reads, such as a web page or email, so that it follows them instead of its user. (https://everyllm.cpn.cleartrust.cc/concepts/prompt-injection) - **Refusal**: When a model declines a request, ideally only for genuinely harmful ones. (https://everyllm.cpn.cleartrust.cc/concepts/refusal) - **Guardrails**: Checks around a model that screen inputs and outputs for harmful, private or off-topic content. (https://everyllm.cpn.cleartrust.cc/concepts/guardrails) - **Sycophancy**: A model's tendency to tell users what they want to hear, agreeing with them even when they are wrong. (https://everyllm.cpn.cleartrust.cc/concepts/sycophancy) - **Reward hacking**: A model finding a way to score well on its training reward without doing what the reward was meant to encourage. (https://everyllm.cpn.cleartrust.cc/concepts/reward-hacking) - **Interpretability**: Research into what happens inside a model: which parts represent which ideas, and how they produce an answer. (https://everyllm.cpn.cleartrust.cc/concepts/interpretability) - **Watermarking**: Hiding a statistical signal in generated text so that it can later be identified as machine-written. (https://everyllm.cpn.cleartrust.cc/concepts/watermarking) - **Model card**: A short document published with a model describing what it is for, how it was trained and evaluated, and its limits. (https://everyllm.cpn.cleartrust.cc/concepts/model-card) - **System card**: A detailed report a lab publishes with a major model, describing its safety testing, risks and safeguards. (https://everyllm.cpn.cleartrust.cc/concepts/system-card) - **Model specification**: A published document setting out how a lab intends its models to behave, and how to weigh conflicting goals. (https://everyllm.cpn.cleartrust.cc/concepts/model-spec) - **Open weights**: A model whose trained weights are published for anyone to download and run, whatever the licence terms. (https://everyllm.cpn.cleartrust.cc/concepts/open-weights) - **Open source AI**: A model released with its weights, code and enough detail on its data that others can study and rebuild it. (https://everyllm.cpn.cleartrust.cc/concepts/open-source-ai) - **Licence**: The legal terms under which a model's weights may be used, changed and shared. (https://everyllm.cpn.cleartrust.cc/concepts/licence) - **Hugging Face Hub**: The largest public site for sharing model weights, datasets and demos, run by the company Hugging Face. (https://everyllm.cpn.cleartrust.cc/concepts/hugging-face-hub) - **GGUF**: A single-file format for quantised models, used by llama.cpp and other tools to run models on ordinary computers. (https://everyllm.cpn.cleartrust.cc/concepts/gguf) - **Model API**: A web service that lets programs send prompts to a hosted model and get answers back, usually paid per token. (https://everyllm.cpn.cleartrust.cc/concepts/api) - **Chatbot**: An application in which people talk with a model in turns, such as ChatGPT, Claude, Gemini or Meta AI. (https://everyllm.cpn.cleartrust.cc/concepts/chatbot) - **Generative pre-trained transformer**: OpenAI's name for its series of decoder models, which has become a general label for models of this kind. (https://everyllm.cpn.cleartrust.cc/concepts/gpt) - **BERT**: Google's 2018 encoder model, which read text in both directions and set new records across language tasks. (https://everyllm.cpn.cleartrust.cc/concepts/bert) - **T5**: Google's 2019 encoder-decoder model that framed every language task as turning one text into another. (https://everyllm.cpn.cleartrust.cc/concepts/t5) - **Multilingual model**: A model trained on many languages so it can understand and write in each, and often translate between them. (https://everyllm.cpn.cleartrust.cc/concepts/multilingual-model) - **Low-resource language**: A language with little digital text available for training, so models handle it less well. (https://everyllm.cpn.cleartrust.cc/concepts/low-resource-language) - **Indic language model**: A model built for the languages of India, such as Hindi, Bengali, Tamil, Telugu and Marathi. (https://everyllm.cpn.cleartrust.cc/concepts/indic-language-model) - **Code-switching**: Mixing two or more languages in one conversation or sentence, as in Hinglish, a common challenge for models. (https://everyllm.cpn.cleartrust.cc/concepts/code-switching) - **Sovereign AI**: A country building its own models and computing capacity, so it does not depend entirely on foreign providers. (https://everyllm.cpn.cleartrust.cc/concepts/sovereign-ai) ## Models - [ChatGPT](https://everyllm.cpn.cleartrust.cc/m/chatgpt) (2022), United States: OpenAI's chatbot built on GPT models, handling text, audio and images - [Grok](https://everyllm.cpn.cleartrust.cc/m/grok) (2023), United States: A chatbot and model line from SpaceX Super Intelligence, launched in 2023. - [Velvet AI](https://everyllm.cpn.cleartrust.cc/m/velvet-ai) (2025), Italy: Italian multilingual open-weight LLMs built for energy efficiency - [Amazon Q](https://everyllm.cpn.cleartrust.cc/m/amazon-q) (2023), United States: Amazon's enterprise chatbot for AWS, with document summaries, data analysis and code. - [Lumo](https://everyllm.cpn.cleartrust.cc/m/lumo-ai-assistant) (2025), Switzerland: A privacy-focused chatbot from Proton with encrypted chat histories - [GLM](https://everyllm.cpn.cleartrust.cc/m/glm) (2021), China: Open-weight Chinese model series used for chat and coding - [GPT-2](https://everyllm.cpn.cleartrust.cc/m/gpt-2) (2019), United States: OpenAI's 2019 text model that predicts the next word in a sequence - [Siri AI](https://everyllm.cpn.cleartrust.cc/m/siri-ai) (2026), United States: Apple's rebuilt Siri assistant, powered by Apple Foundation Models and Google Gemini - [Alice AI](https://everyllm.cpn.cleartrust.cc/m/alice-ai) (2017), Russia: Yandex's Russian voice assistant and generative chatbot, on phones, PCs and smart speakers - [PaLM](https://everyllm.cpn.cleartrust.cc/m/palm) (2022), United States: Google AI's 540-billion-parameter dense decoder-only transformer language model - [Phi](https://everyllm.cpn.cleartrust.cc/m/phi) (2023), United States: Microsoft's small open-weights language models that run on a single device - [GPT-4.1](https://everyllm.cpn.cleartrust.cc/m/gpt-4-1) (2025), United States: Multimodal large language model released by OpenAI in April 2025. - [XLNet](https://everyllm.cpn.cleartrust.cc/m/xlnet) (2019): Permutation-based autoregressive transformer model created by Google AI - [Apertus](https://everyllm.cpn.cleartrust.cc/m/apertus-llm) (2025), Switzerland: A Swiss public large language model released in 8B and 70B sizes - [GPT-6](https://everyllm.cpn.cleartrust.cc/m/gpt-6) (2026), United States: OpenAI's 2026 family of large language models, led by Astra - [BLOOM](https://everyllm.cpn.cleartrust.cc/m/bloom) (2022), France: A 176-billion-parameter multilingual text model built openly by a volunteer research workshop. - [GigaChat](https://everyllm.cpn.cleartrust.cc/m/gigachat) (2023), Russia: Sberbank's Russian-language chatbot that writes text, code and makes images - [Gemini Robotics](https://everyllm.cpn.cleartrust.cc/m/gemini-robotics) (2025), United States: A vision-language-action model from Google DeepMind built for robots - [GloVe](https://everyllm.cpn.cleartrust.cc/m/glove) (2014), United States: Unsupervised word embeddings from global word co-occurrence counts - [Stanford Alpaca](https://everyllm.cpn.cleartrust.cc/m/stanford-alpaca) (2023), United States: A cheap instruction-tuned LLaMA 7B built at Stanford in 2023 - [LaMDA](https://everyllm.cpn.cleartrust.cc/m/lamda) (2020): Google's conversational language model family, first shown in 2021 - [Mistral Vibe](https://everyllm.cpn.cleartrust.cc/m/mistral-vibe) (2025), France: Mistral AI's freemium chatbot for iOS, Android and web. - [Kimi](https://everyllm.cpn.cleartrust.cc/m/kimi) (2023), China: A Chinese chatbot and large language model family by Moonshot AI. - [GPT-5](https://everyllm.cpn.cleartrust.cc/m/gpt-5) (2025), United States: Multimodal large language model released by OpenAI in August 2025. - [DeepSeek](https://everyllm.cpn.cleartrust.cc/m/deepseek) (2025), China: A Chinese series of large language models and a free chatbot app - [GPT-J](https://everyllm.cpn.cleartrust.cc/m/gpt-j) (2021): A 6-billion-parameter open English text generator from EleutherAI, released in 2021. - [ELMo](https://everyllm.cpn.cleartrust.cc/m/elmo) (2018), United States: A 2018 bidirectional LSTM that produced context-sensitive word embeddings. - [Claude Mythos](https://everyllm.cpn.cleartrust.cc/m/claude-mythos) (2026), United States: Anthropic's restricted model family for finding software security flaws - [Sparrow](https://everyllm.cpn.cleartrust.cc/m/sparrow) (2022), United Kingdom: DeepMind's rule-following English chatbot that cites web evidence for its answers - [Whisper](https://everyllm.cpn.cleartrust.cc/m/whisper-speech-recognition-system) (2022): OpenAI's open-source model for turning speech into text and translating it. - [Gemini](https://everyllm.cpn.cleartrust.cc/m/gemini) (2023), United States: Google DeepMind's multimodal LLM family, from Nano to Ultra - [Mamba](https://everyllm.cpn.cleartrust.cc/m/mamba-deep-learning-architecture): Efficient sequence modelling architecture using state spaces - [Gemma](https://everyllm.cpn.cleartrust.cc/m/gemma) (2024), United States: Google DeepMind's open family of Gemini-derived language models - [Wu Dao](https://everyllm.cpn.cleartrust.cc/m/wu-dao) (2021), China: BAAI's trillion-parameter multimodal model family, announced 2021 - [Vicuna](https://everyllm.cpn.cleartrust.cc/m/vicuna-llm) (2023): An open chat model fine-tuned from LLaMA and Llama 2 by LMSYS - [GPT-5.3-Codex](https://everyllm.cpn.cleartrust.cc/m/gpt-5-3-codex) (2026), United States: OpenAI's code-focused large language model with repository and terminal access - [GPT-4](https://everyllm.cpn.cleartrust.cc/m/gpt-4) (2023), United States: Multimodal large language model by OpenAI - [Chinchilla](https://everyllm.cpn.cleartrust.cc/m/chinchilla) (2022), United Kingdom: A 70-billion-parameter DeepMind model built to test how models should scale. - [Amália](https://everyllm.cpn.cleartrust.cc/m/amalia-llm) (2024), Portugal: A Portuguese language model for public administration. - [GPT-5.2](https://everyllm.cpn.cleartrust.cc/m/gpt-5-2) (2025), United States: OpenAI's December 2025 GPT release, offered in three modes plus a coding variant. - [OpenAI o1](https://everyllm.cpn.cleartrust.cc/m/openai-o1) (2024), United States: OpenAI's first reasoning model, which thinks before answering. - [Nemotron](https://everyllm.cpn.cleartrust.cc/m/nemotron) (2023), United States: Nvidia family of language and multimodal models for enterprise and reasoning - [Gato](https://everyllm.cpn.cleartrust.cc/m/gato-deepmind) (2022), United Kingdom: A single DeepMind network handling dialogue, games and robot control - [Jais](https://everyllm.cpn.cleartrust.cc/m/jais) (2023), United Arab Emirates: An open Arabic and English large language model from the UAE - [GPT-5.6](https://everyllm.cpn.cleartrust.cc/m/gpt-5-6) (2026), United States: OpenAI's 2026 model family in Luna, Terra and Sol tiers - [Doubao](https://everyllm.cpn.cleartrust.cc/m/doubao) (2023), China: ByteDance's conversational virtual assistant and language model family - [Inkling](https://everyllm.cpn.cleartrust.cc/m/inkling-large-language-model) (2026): An open-weights mixture-of-experts model handling text, images and audio input - [Microsoft Copilot](https://everyllm.cpn.cleartrust.cc/m/microsoft-copilot) (2023), United States: Microsoft's chatbot assistant built on OpenAI models, replacing Cortana across Windows and Microsoft 365. - [GPT-5.4](https://everyllm.cpn.cleartrust.cc/m/gpt-5-4) (2026): OpenAI's March 2026 large language model with computer use and deep research. - [1.58-bit large language model](https://everyllm.cpn.cleartrust.cc/m/1-58-bit-large-language-model) (2024): A model whose weights hold only three values: -1, 0 and +1. - [Qwen](https://everyllm.cpn.cleartrust.cc/m/qwen) (2023), China: A family of large and small language models developed by Alibaba Cloud. - [OpenAI Codex](https://everyllm.cpn.cleartrust.cc/m/openai-codex) (2025), United States: A code-writing language model fine-tuned from GPT-3 - [BERT](https://everyllm.cpn.cleartrust.cc/m/bert) (2018): An encoder-only transformer that reads text in both directions at once. - [GPT-1](https://everyllm.cpn.cleartrust.cc/m/gpt-1) (2018): OpenAI's first GPT model, a 117M-parameter decoder-only transformer from 2018. - [Alice AI](https://everyllm.cpn.cleartrust.cc/m/alice-ai-ai-model-family) (2023), Russia: Yandex's family of language, vision and image models behind the Alice assistant - [Claude](https://everyllm.cpn.cleartrust.cc/m/claude) (2023), United States: Anthropic's family of large language models and chatbot, from Haiku to Mythos - [Llama](https://everyllm.cpn.cleartrust.cc/m/llama) (2023), United States: Meta AI's family of open-weight language models, from 1B to 2T parameters - [GPT-4chan](https://everyllm.cpn.cleartrust.cc/m/gpt-4chan) (2022): A GPT-J model fine-tuned on posts from 4chan's /pol/ board. - [VideoPoet](https://everyllm.cpn.cleartrust.cc/m/videopoet) (2023), United States: A Google Research language model that turns text and images into video and audio - [Huawei PanGu](https://everyllm.cpn.cleartrust.cc/m/huawei-pangu) (2023), China: Huawei's multimodal large model family for industry, from 1B to trillion-scale - [Minerva](https://everyllm.cpn.cleartrust.cc/m/minerva) (2024), Italy: An Italian-first large language model trained from scratch at Sapienza University of Rome. - [DBRX](https://everyllm.cpn.cleartrust.cc/m/dbrx) (2024): A 132-billion-parameter mixture-of-experts language model from Databricks. - [GPT-4.5](https://everyllm.cpn.cleartrust.cc/m/gpt-4-5) (2025), United States: A large language model from OpenAI, released in 2025 and retired in 2026. - [GPT-3](https://everyllm.cpn.cleartrust.cc/m/gpt-3) (2020), United States: A large language model with 175 billion parameters released by OpenAI in 2020. - [Brave Leo](https://everyllm.cpn.cleartrust.cc/m/brave-leo) (2023), United States: A privacy-focused chatbot built into the Brave browser - [GPT-6.1](https://everyllm.cpn.cleartrust.cc/m/gpt-6-1) (2026), United States: OpenAI's GPT-6.1 family: Sol, Terra and the unreleased Astra models. - [Muse](https://everyllm.cpn.cleartrust.cc/m/muse) (2026), United States: Meta's multimodal LLM family, successor to Llama, with open and closed variants - [Apple Intelligence](https://everyllm.cpn.cleartrust.cc/m/apple-intelligence) (2024), United States: Apple's on-device and cloud AI features built into iOS, iPadOS and macOS - [Manus](https://everyllm.cpn.cleartrust.cc/m/manus-ai-agent) (2025), China: An autonomous AI agent that carries out tasks for you - [Kruti](https://everyllm.cpn.cleartrust.cc/m/kruti) (2025), India: A multilingual agentic chatbot that handles real-world tasks on smartphones. - [Evo](https://everyllm.cpn.cleartrust.cc/m/evo) (2024), United States: Open genomic language models that read and write DNA, RNA and protein sequences - [O3](https://everyllm.cpn.cleartrust.cc/m/openai-o3) (2025), United States: Reasoning model from OpenAI designed for step-by-step logical tasks. - [Tay](https://everyllm.cpn.cleartrust.cc/m/tay) (2016), United States: A Microsoft Twitter chatbot shut down after 16 hours of offensive posts - [Talkie](https://everyllm.cpn.cleartrust.cc/m/talkie) (2026), United States: A 13B small language model trained only on pre-1931 public domain English text - [GPT-5.1](https://everyllm.cpn.cleartrust.cc/m/gpt-5-1) (2025), United States: A five-model GPT family with instant, reasoning and coding variants - [Ernie Bot](https://everyllm.cpn.cleartrust.cc/m/ernie-bot) (2023), China: Baidu chatbot built on ERNIE language models - [T5](https://everyllm.cpn.cleartrust.cc/m/t5) (2019), United States: Encoder-decoder language model family from Google AI using a text-to-text framework. - [OpenAI o4-mini](https://everyllm.cpn.cleartrust.cc/m/openai-o4-mini) (2025), United States: Multimodal reasoning model by OpenAI, now retired. - [BuddhaBot](https://everyllm.cpn.cleartrust.cc/m/buddhabot) (2021), Japan: A chatbot answering questions from early Buddhist texts - [Character.ai](https://everyllm.cpn.cleartrust.cc/m/character-ai) (2022), United States: A chatbot service where users talk to custom-made characters - [HKChat](https://everyllm.cpn.cleartrust.cc/m/hkchat) (2025), Hong Kong: A Hong Kong-focused chatbot app built on the HKGAI V3 language model. - [IBM Granite](https://everyllm.cpn.cleartrust.cc/m/ibm-granite) (2023), United States: IBM's decoder-only foundation models, from 3 to 34 billion parameters - [DALL-E](https://everyllm.cpn.cleartrust.cc/m/dall-e) (2021), United States: Text-to-image model series by OpenAI - [Phi-3](https://everyllm.cpn.cleartrust.cc/m/phi-3) (2024), United States: Small Microsoft language models built to run well on phones and laptops. - [Phi-4-mini](https://everyllm.cpn.cleartrust.cc/m/phi-4-mini) (2025), United States: A 3.8B-parameter compact language model from Microsoft for reasoning, maths and code - [Phi-4-multimodal](https://everyllm.cpn.cleartrust.cc/m/phi-4-multimodal) (2025), United States: A small open model that reads text, images and audio and replies in text - [Phi-4-reasoning](https://everyllm.cpn.cleartrust.cc/m/phi-4-reasoning) (2025), United States: A 14B open-weight reasoning model fine-tuned from Phi-4 - [Falcon2-11B](https://everyllm.cpn.cleartrust.cc/m/falcon-2-11b) (2024), United Arab Emirates: An 11 billion parameter causal decoder model by TII. - [Megatron-LM](https://everyllm.cpn.cleartrust.cc/m/megatron-lm) (2019), United States: Model-parallel training techniques for transformer language models with billions of parameters - [DeepSeek-V3](https://everyllm.cpn.cleartrust.cc/m/deepseek-v3) (2024), China: Open-weight mixture of experts language model with 671 billion parameters. - [Codestral](https://everyllm.cpn.cleartrust.cc/m/codestral), France: Mistral AI's 22B open-weight model built for code generation across 80+ languages - [Codestral Mamba](https://everyllm.cpn.cleartrust.cc/m/codestral-mamba) (2024), France: A 7B open code model built on the Mamba2 architecture - [Falcon 3](https://everyllm.cpn.cleartrust.cc/m/falcon-3) (2024), United Arab Emirates: Open foundation models from 1B to 10B parameters by TII - [Qwen3-Omni](https://everyllm.cpn.cleartrust.cc/m/qwen3-omni) (2025), China: Multimodal model processing text, audio, video and images with real-time speech output. - [Kimi K2 Thinking](https://everyllm.cpn.cleartrust.cc/m/kimi-k2-thinking), China: Open-source reasoning model with tool use and INT4 quantization - [Qwen2.5-Max](https://everyllm.cpn.cleartrust.cc/m/qwen2-5-max) (2025), China: Large-scale mixture of experts model by Alibaba Cloud - [GLM-4.6](https://everyllm.cpn.cleartrust.cc/m/glm-4-6), China: A large language model with a 200K token context window. - [MiniMax-M3](https://everyllm.cpn.cleartrust.cc/m/minimax-m3): Native multimodal model with 1M context and sparse attention - [ERNIE 5.5 Flash](https://everyllm.cpn.cleartrust.cc/m/ernie-5-5-flash), China: Efficient sparse MoE language model with 280B total parameters - [Seed1.5-VL](https://everyllm.cpn.cleartrust.cc/m/seed1-5-vl) (2025): Multimodal vision-language model with reasoning capabilities - [MiMo-V2-Flash](https://everyllm.cpn.cleartrust.cc/m/mimo-v2-flash), China: High-speed MoE language model for reasoning and agents - [MiniCPM-V](https://everyllm.cpn.cleartrust.cc/m/minicpm-v) (2024), China: Efficient multimodal model for mobile devices - [Sarvam-M](https://everyllm.cpn.cleartrust.cc/m/sarvam-m) (2025), India: Hybrid reasoning model for Indian languages and English - [Yi-34B](https://everyllm.cpn.cleartrust.cc/m/yi-34b) (2023), China: A 34-billion parameter bilingual language model by 01.AI - [Falcon 180B](https://everyllm.cpn.cleartrust.cc/m/falcon-180b) (2023), United Arab Emirates: Open 180 billion parameter language model from TII - [AceGPT](https://everyllm.cpn.cleartrust.cc/m/acegpt) (2023), China: An Arabic-focused large language model designed for cultural alignment and local values. - [Project Indus](https://everyllm.cpn.cleartrust.cc/m/project-indus), India: Open-source language model for Hindi and dialects - [Swallow](https://everyllm.cpn.cleartrust.cc/m/swallow) (2024), Japan: Japanese-enhanced Llama models by Tokyo Tech - [SEA-LION](https://everyllm.cpn.cleartrust.cc/m/sea-lion) (2025), Singapore: Multilingual language models for Southeast Asia - [Sarashina2](https://everyllm.cpn.cleartrust.cc/m/sarashina2), Japan: Open-weight Japanese and English language models by SB Intuitions - [SOLAR 10.7B](https://everyllm.cpn.cleartrust.cc/m/solar-10-7b) (2023), South Korea: An 11-billion parameter language model using depth up-scaling. - [RakutenAI-7B](https://everyllm.cpn.cleartrust.cc/m/rakutenai-7b) (2024), Japan: Japanese-focused open language model based on Mistral - [Kimi k1.5](https://everyllm.cpn.cleartrust.cc/m/kimi-k1-5) (2025), China: Multimodal LLM trained with reinforcement learning for reasoning - [PLLuM](https://everyllm.cpn.cleartrust.cc/m/pllum) (2024), Poland: Polish-focused large language model family - [T0](https://everyllm.cpn.cleartrust.cc/m/t0) (2021), France: Encoder-decoder model for zero-shot task generalisation via prompts - [Mercury](https://everyllm.cpn.cleartrust.cc/m/mercury) (2025), United States: Diffusion-based language models for fast code generation - [Snowflake Arctic](https://everyllm.cpn.cleartrust.cc/m/snowflake-arctic) (2024), United States: Open enterprise language model with hybrid MoE design - [StarCoder2](https://everyllm.cpn.cleartrust.cc/m/starcoder2) (2024): Open code generation model by BigCode - [Pythia](https://everyllm.cpn.cleartrust.cc/m/pythia) (2023), United States: A suite of 16 models for interpretability research. - [BioBERT](https://everyllm.cpn.cleartrust.cc/m/biobert) (2019): Biomedical language model based on BERT architecture - [ULMFiT](https://everyllm.cpn.cleartrust.cc/m/ulmfit) (2018): Transfer learning method for NLP text classification tasks - [INTELLECT-1](https://everyllm.cpn.cleartrust.cc/m/intellect-1) (2024): A 10 billion parameter model trained by a global community. - [ChatGLM-6B](https://everyllm.cpn.cleartrust.cc/m/chatglm-6b), China: Open bilingual dialogue model with 6.2 billion parameters - [Jurassic-1](https://everyllm.cpn.cleartrust.cc/m/jurassic-1): AI21 Labs' first widely released language models, in two sizes - [Jurassic-2](https://everyllm.cpn.cleartrust.cc/m/jurassic-2) (2023), Israel: AI21 Labs' second-generation family of large language models in three sizes - [DialoGPT](https://everyllm.cpn.cleartrust.cc/m/dialogpt) (2019), United States: A GPT-style model trained on Reddit threads to write chat replies - [Mixtral 8x7B](https://everyllm.cpn.cleartrust.cc/m/mixtral-8x7b) (2023), France: Sparse mixture-of-experts language model with open weights - [Mistral Saba](https://everyllm.cpn.cleartrust.cc/m/mistral-saba) (2025), France: A 24B regional language model for Arabic and South Asian languages - [Nemotron Nano 2](https://everyllm.cpn.cleartrust.cc/m/nemotron-nano-2) (2025), United States: A 9B hybrid Mamba-Transformer model for reasoning and chat, from NVIDIA. - [Nemotron 3 Super](https://everyllm.cpn.cleartrust.cc/m/nemotron-3-super) (2026), United States: NVIDIA's 120B hybrid MoE language model with 12B active parameters and 1M-token context - [Mistral 7B](https://everyllm.cpn.cleartrust.cc/m/mistral-7b) (2023), France: A 7.3 billion parameter open language model from Mistral AI - [Megatron-Turing NLG 530B](https://everyllm.cpn.cleartrust.cc/m/megatron-turing-nlg-530b) (2022), United States: A 530-billion-parameter generative language model trained by Microsoft and NVIDIA. - [Turing-NLG](https://everyllm.cpn.cleartrust.cc/m/turing-nlg) (2020), United States: Microsoft's 17-billion-parameter Transformer model for generation, QA and summarisation - [Mistral Large](https://everyllm.cpn.cleartrust.cc/m/mistral-large) (2024), France: Mistral AI's flagship multilingual text model, released in 2024 and now deprecated. - [DeBERTa](https://everyllm.cpn.cleartrust.cc/m/deberta) (2020), United States: A BERT-style encoder with disentangled attention and an enhanced mask decoder. - [GLM-130B](https://everyllm.cpn.cleartrust.cc/m/glm-130b) (2022), China: Open bilingual pre-trained language model with 130 billion parameters - [Nemotron-4 340B](https://everyllm.cpn.cleartrust.cc/m/nemotron-4-340b) (2024), United States: NVIDIA's 340B open-access model family for chat and synthetic data generation - [IndicTrans2](https://everyllm.cpn.cleartrust.cc/m/indictrans2) (2023), India: Machine translation for 22 scheduled Indian languages - [IndicBERT](https://everyllm.cpn.cleartrust.cc/m/indicbert) (2022), India: Multilingual encoder model for Indian languages - [IndicBART](https://everyllm.cpn.cleartrust.cc/m/indicbart) (2021), India: Sequence-to-sequence model for Indic languages - [Mistral Medium 3](https://everyllm.cpn.cleartrust.cc/m/mistral-medium-3) (2025), France: A mid-size language model aimed at enterprise coding and multimodal work - [Llama-3.1-Nemotron-Ultra-253B](https://everyllm.cpn.cleartrust.cc/m/llama-3-1-nemotron-ultra-253b), United States: A 253B reasoning model derived from Llama-3.1-405B-Instruct - [BioGPT](https://everyllm.cpn.cleartrust.cc/m/biogpt) (2022), United States: A generative Transformer trained on biomedical writing for text and mining tasks - [Tamil-LLaMA](https://everyllm.cpn.cleartrust.cc/m/tamil-llama) (2023): Open Tamil language model based on LLaMA 2 - [Param-2-17B](https://everyllm.cpn.cleartrust.cc/m/param2-17b-a2-4b-thinking), India: Multilingual MoE model for Indian languages and reasoning - [GLM-4](https://everyllm.cpn.cleartrust.cc/m/glm-4) (2024), China: Multilingual large language model series by Zhipu AI - [OpenHathi](https://everyllm.cpn.cleartrust.cc/m/openhathi), India: A 7 billion parameter language model for Hindi and English. - [Magistral](https://everyllm.cpn.cleartrust.cc/m/magistral) (2025), France: Mistral AI's first reasoning model, in open and enterprise versions - [Mistral Medium 3.5](https://everyllm.cpn.cleartrust.cc/m/mistral-medium-3-5), France: A 128B multimodal model with a 256k window for chat, reasoning and coding - [DeepSeek LLM 67B](https://everyllm.cpn.cleartrust.cc/m/deepseek-llm-67b) (2024), China: A 67-billion-parameter open-source model trained on English and Chinese text. - [Devstral](https://everyllm.cpn.cleartrust.cc/m/devstral) (2025), France: A 24B agentic coding model built by Mistral AI and All Hands AI - [DeepSeekMoE](https://everyllm.cpn.cleartrust.cc/m/deepseekmoe) (2024), China: Mixture-of-experts language model architecture by DeepSeek AI - [SUTRA](https://everyllm.cpn.cleartrust.cc/m/sutra) (2024): Multilingual model separating concepts from language processing - [Ganga-1B](https://everyllm.cpn.cleartrust.cc/m/ganga-1b), India: Hindi-focused autoregressive language model from IIT Gandhinagar - [CodeBERT](https://everyllm.cpn.cleartrust.cc/m/codebert) (2020), China: Bimodal pre-trained model for code and text - [BLIP-2](https://everyllm.cpn.cleartrust.cc/m/blip-2) (2023): Efficient vision-language model using frozen encoders - [OpenCoder](https://everyllm.cpn.cleartrust.cc/m/opencoder) (2024): Open code model family with full training data and pipeline - [SaulLM-7B](https://everyllm.cpn.cleartrust.cc/m/saullm-7b) (2024): A 7-billion parameter language model for legal text. - [BloombergGPT](https://everyllm.cpn.cleartrust.cc/m/bloomberggpt) (2023), United States: A 50 billion parameter language model for finance. - [S1](https://everyllm.cpn.cleartrust.cc/m/s1) (2025), United States: A reasoning model using test-time scaling and budget forcing - [Meditron](https://everyllm.cpn.cleartrust.cc/m/meditron) (2023): Open-source medical language models based on Llama-2. - [Sentence-BERT](https://everyllm.cpn.cleartrust.cc/m/sentence-bert) (2019), Germany: Efficient sentence embeddings using Siamese BERT networks - [Nomic Embed](https://everyllm.cpn.cleartrust.cc/m/nomic-embed) (2024), United States: Open-source long-context text embedding model - [Jina Embeddings v3](https://everyllm.cpn.cleartrust.cc/m/jina-embeddings-v3) (2024), Germany: Multilingual text embedding model with task-specific adapters - [DeepSeekMath](https://everyllm.cpn.cleartrust.cc/m/deepseekmath) (2024), China: Open language model for mathematical reasoning - [DeepSeek-Coder-V2](https://everyllm.cpn.cleartrust.cc/m/deepseek-coder-v2) (2024), China: Open-source mixture-of-experts model for code and math - [DeepSeek-VL2](https://everyllm.cpn.cleartrust.cc/m/deepseek-vl2) (2024), China: Mixture-of-experts vision-language model series for multimodal tasks - [RWKV-7 Goose](https://everyllm.cpn.cleartrust.cc/m/rwkv-7-goose) (2025): Recurrent language model with constant memory usage - [Zamba](https://everyllm.cpn.cleartrust.cc/m/zamba) (2024): A 7-billion parameter hybrid language model combining Mamba and transformer layers. - [Teuken-7B](https://everyllm.cpn.cleartrust.cc/m/teuken-7b) (2024), Germany: Multilingual LLM for all 24 official EU languages - [LFM2](https://everyllm.cpn.cleartrust.cc/m/lfm2), United States: Hybrid language models for edge and on-device use - [TinyLlama](https://everyllm.cpn.cleartrust.cc/m/tinyllama) (2023): A compact 1.1 billion parameter open language model. - [Cerebras-GPT](https://everyllm.cpn.cleartrust.cc/m/cerebras-gpt) (2023), United States: Open compute-optimal language models from Cerebras Systems - [LLaDA](https://everyllm.cpn.cleartrust.cc/m/llada) (2025): An 8-billion parameter diffusion language model - [Eagle and Finch](https://everyllm.cpn.cleartrust.cc/m/eagle-and-finch-rwkv-5-and-rwkv-6) (2024): Efficient recurrent sequence models with matrix states - [StableLM 2](https://everyllm.cpn.cleartrust.cc/m/stablelm-2) (2024), United Kingdom: A small open-weight language model by Stability AI - [INTELLECT-3](https://everyllm.cpn.cleartrust.cc/m/intellect-3): Open 106B MoE reasoning model from Prime Intellect - [Trinity Large](https://everyllm.cpn.cleartrust.cc/m/trinity-large) (2026), United States: 398B-parameter sparse MoE model family by Arcee AI - [Janus-Pro](https://everyllm.cpn.cleartrust.cc/m/janus-pro) (2025), China: Unified multimodal model for understanding and generating images - [Hermes 3](https://everyllm.cpn.cleartrust.cc/m/hermes-3) (2024), United States: Open-weight instruct model based on Llama-3.1 405B - [DeepSeek-V3.1](https://everyllm.cpn.cleartrust.cc/m/deepseek-v3-1), China: Hybrid thinking and non-thinking language model with 671 billion parameters. - [DeepSeek-Prover-V2](https://everyllm.cpn.cleartrust.cc/m/deepseek-prover-v2) (2025), China: Open-source model for formal theorem proving in Lean 4 - [GPT-NeoX-20B](https://everyllm.cpn.cleartrust.cc/m/gpt-neox-20b) (2022), United States: Open 20 billion parameter autoregressive language model by EleutherAI - [DeepSeek-V3.2](https://everyllm.cpn.cleartrust.cc/m/deepseek-v3-2) (2025), China: Efficient reasoning and agentic AI model with sparse attention - [DeepSeek-V4-Pro](https://everyllm.cpn.cleartrust.cc/m/deepseek-v4-pro) (2026), China: 1.6T parameter MoE model with one million token context - [SmolLM3](https://everyllm.cpn.cleartrust.cc/m/smollm3), United States: Open 3B parameter language model with reasoning modes - [StarCoder](https://everyllm.cpn.cleartrust.cc/m/starcoder) (2023): Open code generation model with 15.5 billion parameters - [SmolLM2](https://everyllm.cpn.cleartrust.cc/m/smollm2) (2025), United States: Compact language models for on-device use - [Zephyr 7B](https://everyllm.cpn.cleartrust.cc/m/zephyr-7b) (2023), United States: A 7 billion parameter chat model fine-tuned for helpfulness. - [DistilBERT](https://everyllm.cpn.cleartrust.cc/m/distilbert) (2019): A smaller, faster version of the BERT language model. - [GPT-Neo](https://everyllm.cpn.cleartrust.cc/m/gpt-neo): Open autoregressive language model by EleutherAI - [Llemma](https://everyllm.cpn.cleartrust.cc/m/llemma) (2023), United States: Open language model for mathematical reasoning - [ALIA-40b](https://everyllm.cpn.cleartrust.cc/m/alia-40b), Spain: Multilingual base model for 35 European languages - [OLMo 3](https://everyllm.cpn.cleartrust.cc/m/olmo-3) (2025), United States: Open language models for scientific research by Allen Institute for AI - [Tülu 3](https://everyllm.cpn.cleartrust.cc/m/tulu-3) (2024), United States: Open post-trained model family built on Llama 3.1 - [OLMo 2](https://everyllm.cpn.cleartrust.cc/m/olmo-2) (2024), United States: Open language model family with full training transparency - [SciBERT](https://everyllm.cpn.cleartrust.cc/m/scibert) (2019), United States: A BERT model trained on scientific text. - [DeepSeek-V4-Flash](https://everyllm.cpn.cleartrust.cc/m/deepseek-v4-flash) (2026), China: Efficient mixture-of-experts model with one million token context - [DeepSeek-V4.1-Flash](https://everyllm.cpn.cleartrust.cc/m/deepseek-v4-1-flash), China: Multimodal MoE model with efficient KV cache compression - [Qwen-72B](https://everyllm.cpn.cleartrust.cc/m/qwen-72b) (2023), China: A 72-billion parameter open-weight language model from Alibaba Cloud. - [Molmo](https://everyllm.cpn.cleartrust.cc/m/molmo) (2024), United States: Open vision-language model family by Allen Institute for AI - [Qwen1.5](https://everyllm.cpn.cleartrust.cc/m/qwen1-5) (2024), China: Open-weight language model series with nine sizes and strong multilingual support. - [K-EXAONE 2.0](https://everyllm.cpn.cleartrust.cc/m/k-exaone-2-0), South Korea: Open-weight multilingual language model by LG AI Research - [Pharia-1-LLM-7B](https://everyllm.cpn.cleartrust.cc/m/pharia-1-llm), Germany: Multilingual 7B parameter decoder model by Aleph Alpha Research - [Poro 2](https://everyllm.cpn.cleartrust.cc/m/poro-2), Finland: Finnish and English instruction-following model based on Llama 3.1 - [PhoGPT](https://everyllm.cpn.cleartrust.cc/m/phogpt) (2023), Vietnam: Open-source generative model for Vietnamese language tasks. - [Lucie-7B](https://everyllm.cpn.cleartrust.cc/m/lucie-7b) (2025), France: Open multilingual language model focused on French and English - [CroissantLLM](https://everyllm.cpn.cleartrust.cc/m/croissantllm) (2024), France: A 1.3 billion parameter bilingual French-English language model. - [Bielik](https://everyllm.cpn.cleartrust.cc/m/bielik) (2025), Poland: Polish-optimised 11 billion parameter language model - [Moshi](https://everyllm.cpn.cleartrust.cc/m/moshi) (2024), France: Real-time full-duplex spoken dialogue model - [BLOOMZ](https://everyllm.cpn.cleartrust.cc/m/bloomz) (2022): Multilingual instruction-following model based on BLOOM - [CamemBERT](https://everyllm.cpn.cleartrust.cc/m/camembert) (2019), France: French language model based on RoBERTa architecture - [Qwen2.5-VL](https://everyllm.cpn.cleartrust.cc/m/qwen2-5-vl) (2025), China: Vision-language model for images, documents and long videos - [QwQ-32B](https://everyllm.cpn.cleartrust.cc/m/qwq-32b) (2025), China: A 32-billion parameter reasoning model from the Qwen series. - [Qwen3-235B-A22B](https://everyllm.cpn.cleartrust.cc/m/qwen3-235b-a22b) (2025), China: Mixture-of-experts model with switchable reasoning modes - [PLaMo-100B](https://everyllm.cpn.cleartrust.cc/m/plamo-100b) (2024), Japan: A 100-billion parameter language model built for Japanese. - [A.X K1](https://everyllm.cpn.cleartrust.cc/m/a-x-k1) (2026), South Korea: Sparse MoE model with controllable reasoning depth - [SeaLLMs](https://everyllm.cpn.cleartrust.cc/m/seallm) (2023), Singapore: Open language models for Southeast Asian languages - [EvoLLM-JP](https://everyllm.cpn.cleartrust.cc/m/evollm-jp) (2024), Japan: Experimental Japanese language model created by merging open-source models. - [Motif-2](https://everyllm.cpn.cleartrust.cc/m/motif-2): A 12.7 billion parameter reasoning model by Motif Technologies. - [MaLLaM](https://everyllm.cpn.cleartrust.cc/m/mallam) (2024), Malaysia: Large language models trained on Malaysian context and Malay language data. - [Taiwan-LLM](https://everyllm.cpn.cleartrust.cc/m/taiwan-llm) (2023), Taiwan: Open language model for Traditional Chinese in Taiwan - [PLaMo 2](https://everyllm.cpn.cleartrust.cc/m/plamo-2) (2025), Japan: Japanese-focused large language model series by Preferred Networks - [RuGPT-3](https://everyllm.cpn.cleartrust.cc/m/rugpt-3), Russia: Russian language model based on GPT-2 architecture - [YaLM 100B](https://everyllm.cpn.cleartrust.cc/m/yalm-100b), Russia: Open-weight Russian and English language model with 100 billion parameters - [Qwen3-Coder](https://everyllm.cpn.cleartrust.cc/m/qwen3-coder) (2025), China: Agentic coding model with 480 billion parameters - [Qwen3-Next-80B-A3B](https://everyllm.cpn.cleartrust.cc/m/qwen3-next-80b-a3b), China: An 80-billion parameter hybrid model with ultra-long context support. - [Qwen3-VL](https://everyllm.cpn.cleartrust.cc/m/qwen3-vl), China: Vision-language model with deep visual reasoning and long video context. - [Qwen3.5-397B-A17B](https://everyllm.cpn.cleartrust.cc/m/qwen3-5-397b-a17b), China: A large multimodal language model with sparse activation. - [Qwen3.6-35B-A3B](https://everyllm.cpn.cleartrust.cc/m/qwen3-6-35b-a3b) (2026), China: Open-weight multimodal coding model with hybrid architecture - [HyperCLOVA X SEED Think](https://everyllm.cpn.cleartrust.cc/m/hyperclova-x-seed-think), South Korea: Efficient reasoning model for Korean and multimodal tasks - [HyperCLOVA X](https://everyllm.cpn.cleartrust.cc/m/hyperclova-x) (2024), South Korea: Korean-focused large language model with English and coding skills - [EXAONE 3.0](https://everyllm.cpn.cleartrust.cc/m/exaone-3-0) (2024), South Korea: Open bilingual language model from LG AI Research - [Tsuzumi](https://everyllm.cpn.cleartrust.cc/m/tsuzumi) (2024), Japan: Lightweight Japanese LLM for business use - [EXAONE 3.5](https://everyllm.cpn.cleartrust.cc/m/exaone-3-5) (2024), South Korea: LG AI Research open language model series for real-world tasks - [EXAONE 4.0](https://everyllm.cpn.cleartrust.cc/m/exaone-4-0) (2025), South Korea: Unified language model with reasoning and non-reasoning modes - [EXAONE 4.5](https://everyllm.cpn.cleartrust.cc/m/exaone-4-5), South Korea: Open-weight vision-language model from LG AI Research - [PanGu-Alpha](https://everyllm.cpn.cleartrust.cc/m/pangu-alpha) (2021), China: A large Chinese language model with 200 billion parameters. - [Solar Open 100B](https://everyllm.cpn.cleartrust.cc/m/solar-open-100b) (2026), South Korea: 102B parameter Mixture-of-Experts model by Upstage - [Kanana](https://everyllm.cpn.cleartrust.cc/m/kanana) (2025), South Korea: Bilingual language models for Korean and English by Kakao. - [Solar Open 2](https://everyllm.cpn.cleartrust.cc/m/solar-open2-250b) (2026), South Korea: 250B parameter open-weight model for agentic tasks - [EuroLLM](https://everyllm.cpn.cleartrust.cc/m/eurollm) (2024): Open multilingual models for European languages - [Kanana-2](https://everyllm.cpn.cleartrust.cc/m/kanana-2), South Korea: Open-source multilingual model for agentic AI and reasoning. - [Qwen3.8-2.4T-A95B](https://everyllm.cpn.cleartrust.cc/m/qwen3-8-2-4t-a95b) (2026), China: Open-weight reasoning model with 2.4 trillion parameters - [Kimi K2.6](https://everyllm.cpn.cleartrust.cc/m/kimi-k2-6), China: Open-source multimodal agentic model by Moonshot AI - [Falcon-H1R-7B](https://everyllm.cpn.cleartrust.cc/m/falcon-h1r) (2026), United Arab Emirates: A 7-billion parameter reasoning model with hybrid architecture - [Kimi K2](https://everyllm.cpn.cleartrust.cc/m/kimi-k2) (2025), China: Open mixture-of-experts model for agentic tasks and coding - [Falcon-H1](https://everyllm.cpn.cleartrust.cc/m/falcon-h1) (2025), United Arab Emirates: Hybrid transformer and state-space model series by TII - [Fanar](https://everyllm.cpn.cleartrust.cc/m/fanar) (2025), Qatar: Arabic-centric multimodal generative AI platform and models - [Seed Diffusion Preview](https://everyllm.cpn.cleartrust.cc/m/seed-diffusion-preview) (2025): Fast discrete diffusion language model for code - [Falcon-40B](https://everyllm.cpn.cleartrust.cc/m/falcon-40b) (2023), United Arab Emirates: A 40-billion parameter open language model from TII. - [Sherkala-Chat](https://everyllm.cpn.cleartrust.cc/m/sherkala) (2025), Kazakhstan: Open LLM for Kazakh language tasks - [LLM-jp-3](https://everyllm.cpn.cleartrust.cc/m/llm-jp-3) (2024), Japan: Open Japanese language model family by NII - [Llama-3-ELYZA-JP-8B](https://everyllm.cpn.cleartrust.cc/m/elyza-jp-8b), Japan: Japanese-tuned Llama 3 model by ELYZA - [Stockmark-100b](https://everyllm.cpn.cleartrust.cc/m/stockmark-100b), Japan: A 100 billion parameter Japanese and English language model. - [Kimi K3.1 Pro](https://everyllm.cpn.cleartrust.cc/m/kimi-k3-1-pro), China: Open-weight multimodal agentic model with 1M token context - [GLM-4.5](https://everyllm.cpn.cleartrust.cc/m/glm-4-5) (2025), China: Open-source mixture of experts model for agents and reasoning - [Kimi Linear](https://everyllm.cpn.cleartrust.cc/m/kimi-linear) (2025), China: Hybrid linear attention model with 48 billion parameters - [Navarasa 2.0](https://everyllm.cpn.cleartrust.cc/m/navarasa), India: Instruction-tuned model for 15 Indian languages and English. - [Fathom-R1-14B](https://everyllm.cpn.cleartrust.cc/m/fathom-r1-14b): Low-cost open reasoning model for mathematics - [Ling-1T](https://everyllm.cpn.cleartrust.cc/m/ling-1t), China: Trillion-parameter efficient reasoning model by InclusionAI - [GLM-4.7](https://everyllm.cpn.cleartrust.cc/m/glm-4-7), China: Coding and reasoning model from Z.ai - [LongCat-2.5](https://everyllm.cpn.cleartrust.cc/m/longcat-2-5), China: Multimodal MoE model for long-horizon agentic tasks - [LongCat-2.0](https://everyllm.cpn.cleartrust.cc/m/longcat-2-0), China: A 1.6 trillion parameter MoE language model from Meituan. - [MiniCPM](https://everyllm.cpn.cleartrust.cc/m/minicpm) (2024), China: Efficient small language models for end-side devices - [Sarvam-1](https://everyllm.cpn.cleartrust.cc/m/sarvam-1) (2024), India: A 2-billion parameter language model for Indian languages. - [Krutrim-1](https://everyllm.cpn.cleartrust.cc/m/krutrim-1) (2024), India: Multilingual language model for Indian languages - [Shuka v1](https://everyllm.cpn.cleartrust.cc/m/shuka-v1), India: Audio-native language model for Indic languages - [Chitrarth](https://everyllm.cpn.cleartrust.cc/m/chitrarth), India: Multilingual vision-language model for ten Indian languages and English. - [Krutrim-2](https://everyllm.cpn.cleartrust.cc/m/krutrim-2) (2024), India: A 12-billion parameter multilingual language model focused on Indian languages and context. - [GLM-5.3](https://everyllm.cpn.cleartrust.cc/m/glm-5-3) (2026), China: Open-weights coding and cyber-security model from Z.ai - [CogVLM](https://everyllm.cpn.cleartrust.cc/m/cogvlm) (2023), China: Open-source visual language model with deep feature fusion - [MiniMax-Text-01](https://everyllm.cpn.cleartrust.cc/m/minimax-text-01) (2025), China: A large language model with a four million token context window. - [MiniMax-M1](https://everyllm.cpn.cleartrust.cc/m/minimax-m1) (2025), China: Open-weight reasoning model with hybrid attention and long context - [Sarvam 2B](https://everyllm.cpn.cleartrust.cc/m/sarvam-2b) (2024), India: Small language model for English and ten Indian languages. - [LongCat-Flash](https://everyllm.cpn.cleartrust.cc/m/longcat-flash) (2025), China: Efficient 560B parameter MoE model for agentic tasks - [Sarvam-30B](https://everyllm.cpn.cleartrust.cc/m/sarvam-30b), India: Open-weight MoE model for Indian languages and coding - [PanGu-Sigma](https://everyllm.cpn.cleartrust.cc/m/pangu-sigma) (2023), China: Trillion-parameter sparse language model for Chinese NLP tasks - [GLM-5](https://everyllm.cpn.cleartrust.cc/m/glm-5) (2026), China: Open-weight model for engineering and agentic tasks - [MiniMax-M2.5](https://everyllm.cpn.cleartrust.cc/m/minimax-m2-5) (2026), China: Cost-efficient coding and agentic model from MiniMax AI - [ERNIE 3.0](https://everyllm.cpn.cleartrust.cc/m/ernie-3-0) (2021), China: Knowledge-enhanced pre-training model for language tasks - [ERNIE 3.0 Titan](https://everyllm.cpn.cleartrust.cc/m/ernie-3-0-titan) (2021), China: Large-scale Chinese knowledge-enhanced pre-trained language model - [ERNIE 1.0](https://everyllm.cpn.cleartrust.cc/m/ernie-1-0) (2019), China: Knowledge-enhanced language model for Chinese NLP tasks. - [ERNIE-4.5-21B-A3B-Thinking](https://everyllm.cpn.cleartrust.cc/m/ernie-4-5-21b-a3b-thinking) (2025), China: Lightweight reasoning model with 21 billion parameters - [ERNIE 4.5](https://everyllm.cpn.cleartrust.cc/m/ernie-4-5) (2025), China: Baidu's multimodal mixture-of-experts language model series. - [Hunyuan-A13B](https://everyllm.cpn.cleartrust.cc/m/hunyuan-a13b) (2025), China: Open-source MoE language model with 80 billion total parameters - [Hunyuan-TurboS](https://everyllm.cpn.cleartrust.cc/m/hunyuan-turbos) (2025), China: Hybrid Mamba-Transformer MoE model by Tencent - [Seed1.5-Thinking](https://everyllm.cpn.cleartrust.cc/m/seed1-5-thinking) (2025), China: A reasoning model from ByteDance Seed using reinforcement learning - [Step3](https://everyllm.cpn.cleartrust.cc/m/step3) (2025), China: Multimodal reasoning model with 321 billion parameters - [Seed-OSS-36B](https://everyllm.cpn.cleartrust.cc/m/seed-oss-36b) (2025), China: Open-source 36B parameter language model with controllable reasoning. - [Intern-S1](https://everyllm.cpn.cleartrust.cc/m/intern-s1) (2025), China: Open-source multimodal reasoning model for science - [MiMo-7B](https://everyllm.cpn.cleartrust.cc/m/mimo-7b) (2025), China: Open reasoning model from Xiaomi for maths and code - [Pangu Ultra](https://everyllm.cpn.cleartrust.cc/m/pangu-ultra) (2025), China: A 135-billion parameter dense language model trained on Ascend hardware. - [InternVL3](https://everyllm.cpn.cleartrust.cc/m/internvl3) (2025), China: Open-source multimodal large language model series - [UI-TARS](https://everyllm.cpn.cleartrust.cc/m/ui-tars) (2025), China: Open-source multimodal agent for GUI tasks - [MiMo-V2.5-Pro](https://everyllm.cpn.cleartrust.cc/m/mimo-v2-5-pro), China: Open-source MoE language model with 1M token context - [Baichuan-M2](https://everyllm.cpn.cleartrust.cc/m/baichuan-m2) (2025), China: Open-source medical reasoning model with 32 billion parameters. - [Baichuan 2](https://everyllm.cpn.cleartrust.cc/m/baichuan-2) (2023), China: Open multilingual language models with 7B and 13B parameters. - [MiMo-V2.6-Pro](https://everyllm.cpn.cleartrust.cc/m/mimo-v2-6-pro), China: Xiaomi’s 1T parameter multimodal model with 1M token context - [Grok Code Fast 1](https://everyllm.cpn.cleartrust.cc/m/grok-code-fast-1) (2026), United States: A fast, low-cost reasoning model built for agentic coding work. - [NVLM 1.0](https://everyllm.cpn.cleartrust.cc/m/nvlm-1-0) (2024), United States: Open multimodal language models from NVIDIA for vision and language tasks - [OpenELM](https://everyllm.cpn.cleartrust.cc/m/openelm) (2024), United States: Apple's small open language models with a full training and inference release - [Devstral 2](https://everyllm.cpn.cleartrust.cc/m/devstral-2) (2026), France: Open-weight coding model family for software agents - [K2-Think](https://everyllm.cpn.cleartrust.cc/m/k2-think) (2025): A 32 billion parameter open reasoning model - [Fugaku-LLM](https://everyllm.cpn.cleartrust.cc/m/fugaku-llm) (2024), Japan: Japanese large language model trained on Fugaku supercomputer - [Step-3.5-Flash](https://everyllm.cpn.cleartrust.cc/m/step-3-5-flash) (2026), China: Open-source MoE model for reasoning and agents - [Qwen2.5-72B](https://everyllm.cpn.cleartrust.cc/m/qwen2-5-72b) (2024), China: A 72-billion parameter open-weight language model from Alibaba Cloud. - [MAI-1-preview](https://everyllm.cpn.cleartrust.cc/m/mai-1-preview) (2025), United States: Microsoft AI's first in-house foundation model, a mixture-of-experts system for everyday queries - [Gemini Diffusion](https://everyllm.cpn.cleartrust.cc/m/gemini-diffusion), United States: Experimental text diffusion model from Google DeepMind - [CodeGen](https://everyllm.cpn.cleartrust.cc/m/codegen) (2022), United States: Open autoregressive language model for code generation - [EXAONE Deep](https://everyllm.cpn.cleartrust.cc/m/exaone-deep) (2025), South Korea: Reasoning-focused language models from LG AI Research - [DeepSeek Coder](https://everyllm.cpn.cleartrust.cc/m/deepseek-coder) (2024), China: Open-source code generation models from DeepSeek - [Mistral Small 3](https://everyllm.cpn.cleartrust.cc/m/mistral-small-3) (2025), France: A 24B open-weight model tuned for low latency and local deployment - [OPT](https://everyllm.cpn.cleartrust.cc/m/opt) (2022), United States: Meta AI's family of open decoder-only language models from 125M to 175B parameters - [Pragna-1B](https://everyllm.cpn.cleartrust.cc/m/pragna-1b), India: Small decoder model for Hindi, Bangla, Gujarati and English - [Typhoon](https://everyllm.cpn.cleartrust.cc/m/typhoon) (2024), Thailand: Thai-focused large language model series by SCB 10X - [Flan-T5](https://everyllm.cpn.cleartrust.cc/m/flan-t5) (2022), United States: Instruction-finetuned T5 checkpoints released by Google for zero-shot and few-shot work - [Llama 3](https://everyllm.cpn.cleartrust.cc/m/llama-3) (2024), United States: Meta's open-weight language model family in 8B and 70B sizes - [Nemotron 3.5 Super VL](https://everyllm.cpn.cleartrust.cc/m/nemotron-3-5-super-vl) (2026), United States: NVIDIA's open-weight vision-language model with 120B parameters and 1M-token context - [Jamba2 Mini](https://everyllm.cpn.cleartrust.cc/m/jamba2): A small open-weights SSM-Transformer model for enterprise work - [Ministral 3B and 8B](https://everyllm.cpn.cleartrust.cc/m/ministral-3b-and-8b) (2024), France: Two small Mistral AI models built for on-device and edge computing. - [DiffusionGemma](https://everyllm.cpn.cleartrust.cc/m/diffusiongemma), United States: A multimodal Gemma 4 model that writes text by denoising blocks of tokens. - [Qwen2](https://everyllm.cpn.cleartrust.cc/m/qwen2) (2024), China: Open-weight language model series from Alibaba - [Mamba-2](https://everyllm.cpn.cleartrust.cc/m/mamba-2) (2024), United States: A state-space model architecture for efficient language modelling - [GLM-4.1V-Thinking](https://everyllm.cpn.cleartrust.cc/m/glm-4-1v-thinking) (2025), China: Open-source vision-language model with reasoning focus - [Qwen2.5-Coder](https://everyllm.cpn.cleartrust.cc/m/qwen2-5-coder) (2024), China: Open code-specialised language model series by Alibaba Cloud - [Sailor](https://everyllm.cpn.cleartrust.cc/m/sailor) (2024), Singapore: Open multilingual language models for South-East Asia - [Gemini 2.5 Pro](https://everyllm.cpn.cleartrust.cc/m/gemini-2-5-pro) (2025), United States: Google DeepMind's thinking model for complex reasoning, code and long-context work - [Sahabat-AI](https://everyllm.cpn.cleartrust.cc/m/sahabat-ai), Indonesia: Indonesian language model family by GoTo and Indosat - [AlphaCode](https://everyllm.cpn.cleartrust.cc/m/alphacode) (2022), United Kingdom: A code generation system that solves competitive programming problems. - [ELECTRA](https://everyllm.cpn.cleartrust.cc/m/electra) (2020): A pre-training method that trains text encoders to spot replaced tokens. - [Claude Opus 4.7](https://everyllm.cpn.cleartrust.cc/m/claude-opus-4-7) (2026), United States: Anthropic's large model for long-running coding and professional work - [Pixtral 12B](https://everyllm.cpn.cleartrust.cc/m/pixtral-12b) (2024), France: A 12B multimodal model that reads images and text without losing text skills - [Mistral NeMo](https://everyllm.cpn.cleartrust.cc/m/mistral-nemo) (2024), France: A 12B multilingual open-weights model built with NVIDIA, with a 128k context window. - [Grok 3](https://everyllm.cpn.cleartrust.cc/m/grok-3) (2025), United States: xAI's reasoning-focused language model, released in beta in 2025 - [Aya Expanse](https://everyllm.cpn.cleartrust.cc/m/aya-expanse) (2024): Open-weight multilingual language models in 8B and 32B sizes from Cohere For AI - [DeepSeek-OCR](https://everyllm.cpn.cleartrust.cc/m/deepseek-ocr) (2025): Optical character recognition model using visual compression - [Zamba2](https://everyllm.cpn.cleartrust.cc/m/zamba2) (2024), United States: Hybrid Mamba2-transformer language model suite by Zyphra - [Poro](https://everyllm.cpn.cleartrust.cc/m/poro) (2024), Finland: Open 34B parameter model for Finnish, English and code - [Grok-1](https://everyllm.cpn.cleartrust.cc/m/grok-1) (2024), United States: xAI's 314 billion parameter open-weights mixture-of-experts base model - [Gemini 2.5 Flash](https://everyllm.cpn.cleartrust.cc/m/gemini-2-5-flash) (2025), United States: A hybrid reasoning model from Google with adjustable thinking effort - [T-Pro 2.0](https://everyllm.cpn.cleartrust.cc/m/t-pro-2-0), Russia: Russian-tuned reasoning model based on Qwen 3 - [Sarvam-105B](https://everyllm.cpn.cleartrust.cc/m/sarvam-105b), India: Open-weight MoE model for Indian languages and reasoning - [Reka Core](https://everyllm.cpn.cleartrust.cc/m/reka-core) (2024), Singapore: Multimodal language model by Reka - [Kimi K2.5](https://everyllm.cpn.cleartrust.cc/m/kimi-k2-5), China: Open-source multimodal agentic model with vision and reasoning. - [Llama 3.3 70B](https://everyllm.cpn.cleartrust.cc/m/llama-3-3-70b), United States: A text-only 70B instruction-tuned model from Meta's Llama 3 line. - [Hunyuan-Large](https://everyllm.cpn.cleartrust.cc/m/hunyuan-large) (2024), China: Open-source mixture of experts model with 389 billion parameters - [Param-1](https://everyllm.cpn.cleartrust.cc/m/param-1), India: Bilingual and multilingual language models from India. - [Salamandra](https://everyllm.cpn.cleartrust.cc/m/salamandra) (2025), Spain: Open multilingual language models for European languages - [RETRO](https://everyllm.cpn.cleartrust.cc/m/retro) (2021): A retrieval-enhanced transformer that conditions on chunks from a huge document database. - [Falcon Mamba 7B](https://everyllm.cpn.cleartrust.cc/m/falcon-mamba-7b) (2024), United Arab Emirates: A 7-billion parameter language model using the Mamba architecture. - [MiniMax-M2](https://everyllm.cpn.cleartrust.cc/m/minimax-m2), China: Open-source MoE model for coding and agents - [Nemotron 3 Nano](https://everyllm.cpn.cleartrust.cc/m/nemotron-3-nano) (2025), United States: NVIDIA's 30B hybrid Mamba-Transformer MoE model for reasoning and agents - [AfriBERTa](https://everyllm.cpn.cleartrust.cc/m/afriberta) (2021): Multilingual encoder model for eleven African languages - [Kimi K3](https://everyllm.cpn.cleartrust.cc/m/kimi-k3) (2026), China: Open-weight multimodal agentic model with 2.8T parameters - [Airavata](https://everyllm.cpn.cleartrust.cc/m/airavata) (2024), India: Instruction-tuned large language model for Hindi - [PaLM 2](https://everyllm.cpn.cleartrust.cc/m/palm-2) (2023), United States: Google's multilingual Transformer language model in four sizes, from on-device to largest - [GPT-SW3](https://everyllm.cpn.cleartrust.cc/m/gpt-sw3) (2023), Sweden: Generative language model for Nordic languages - [Sarvam-Translate](https://everyllm.cpn.cleartrust.cc/m/sarvam-translate), India: Document-level translation model for 22 Indian languages. - [Qwen2.5-Omni](https://everyllm.cpn.cleartrust.cc/m/qwen2-5-omni) (2025), China: End-to-end multimodal model for real-time speech and video chat. - [Toolformer](https://everyllm.cpn.cleartrust.cc/m/toolformer) (2023), United States: A language model that learns on its own when to call external tools through APIs. - [ModernBERT](https://everyllm.cpn.cleartrust.cc/m/modernbert) (2024), United States: A modernised encoder model for retrieval and classification tasks. - [Viking](https://everyllm.cpn.cleartrust.cc/m/viking) (2023), Finland: Open source language model for Nordic languages and code - [Leanstral 1.5](https://everyllm.cpn.cleartrust.cc/m/leanstral-1-5), France: An open code agent model built for the Lean 4 proof assistant - [Granite 3.0](https://everyllm.cpn.cleartrust.cc/m/granite-3-0) (2024), United States: IBM's third-generation open enterprise language models, from 2B to 8B parameters - [Aya 101](https://everyllm.cpn.cleartrust.cc/m/aya-101) (2024): A 13-billion-parameter multilingual model that follows instructions in 101 languages - [Falcon-7B](https://everyllm.cpn.cleartrust.cc/m/falcon-7b) (2023), United Arab Emirates: Open 7-billion parameter decoder model by TII - [Amazon Nova](https://everyllm.cpn.cleartrust.cc/m/amazon-nova) (2024), United States: Amazon's family of foundation models on AWS Bedrock - [North Mini Code](https://everyllm.cpn.cleartrust.cc/m/north-mini-code), United States: A 30B sparse mixture-of-experts model for code and terminal agent work - [Granite 4.2](https://everyllm.cpn.cleartrust.cc/m/granite-4-2) (2026), United States: IBM's open reasoning language model family, released August 2026 - [Nemotron-4-Mini-Hindi-4B-Instruct](https://everyllm.cpn.cleartrust.cc/m/nemotron-4-mini-hindi-4b) (2024), United States: A small Hindi-focused chat model from NVIDIA, tuned for Indian-context answers - [Amazon Nova Premier](https://everyllm.cpn.cleartrust.cc/m/amazon-nova-premier), United States: Amazon's largest Nova model for complex tasks and model distillation - [Galactica](https://everyllm.cpn.cleartrust.cc/m/galactica) (2022), United States: A language model trained on scientific text for research tasks - [Granite 4.0](https://everyllm.cpn.cleartrust.cc/m/granite-4-0) (2025), United States: IBM's hybrid Mamba/transformer language models for enterprise work - [Mistral Large 3](https://everyllm.cpn.cleartrust.cc/m/mistral-large-3), France: A 675B-parameter open-weight mixture-of-experts multimodal model from Mistral AI - [Apple Foundation Models](https://everyllm.cpn.cleartrust.cc/m/apple-foundation-models-2024) (2024), United States: Apple's on-device and server language models behind Apple Intelligence - [MM1](https://everyllm.cpn.cleartrust.cc/m/mm1) (2024): A family of multimodal language models up to 30B parameters, dense and MoE. - [Apple Foundation Models](https://everyllm.cpn.cleartrust.cc/m/apple-foundation-models-2025) (2025), United States: Apple's 2025 on-device and server language models for Apple Intelligence - [AlexaTM 20B](https://everyllm.cpn.cleartrust.cc/m/alexatm-20b) (2022), United States: A 20-billion-parameter multilingual sequence-to-sequence model from Amazon's Alexa team. - [Orca 2](https://everyllm.cpn.cleartrust.cc/m/orca-2) (2023), United States: A small LLaMA-2-based model trained on synthetic data to reason. - [Grok-1.5](https://everyllm.cpn.cleartrust.cc/m/grok-1-5) (2024), United States: xAI's long-context Grok model with stronger reasoning and coding skills - [Grok 4](https://everyllm.cpn.cleartrust.cc/m/grok-4) (2025), United States: xAI's flagship reasoning model with native tool use and live search - [Grok-2](https://everyllm.cpn.cleartrust.cc/m/grok-2) (2024), United States: xAI's 2024 chat and reasoning model, with a smaller sibling, Grok-2 mini - [Command R](https://everyllm.cpn.cleartrust.cc/m/command-r): Cohere's model built for retrieval-augmented generation in production - [Mixtral 8x22B](https://everyllm.cpn.cleartrust.cc/m/mixtral-8x22b), France: Sparse mixture-of-experts open model with 39B active parameters out of 141B - [Grok 4.1](https://everyllm.cpn.cleartrust.cc/m/grok-4-1) (2025), United States: xAI's Grok 4.1, tuned for personality, emotional tone and fewer factual errors - [Command A](https://everyllm.cpn.cleartrust.cc/m/command-a) (2025): Cohere's enterprise-focused large language model for agents, search and tool use - [Command R+](https://everyllm.cpn.cleartrust.cc/m/command-r-plus), Canada: Cohere's large language model built for enterprise workloads and retrieval-augmented use. - [Phi-2](https://everyllm.cpn.cleartrust.cc/m/phi-2), United States: A 2.7 billion parameter Microsoft research model for QA, chat and code - [Aya Vision](https://everyllm.cpn.cleartrust.cc/m/aya-vision): Open-weights multilingual multimodal vision model from Cohere For AI - [InstructGPT](https://everyllm.cpn.cleartrust.cc/m/instructgpt) (2022), United States: GPT-3 models fine-tuned with human feedback to follow instructions - [WebGPT](https://everyllm.cpn.cleartrust.cc/m/webgpt) (2021), United States: A GPT-3 model fine-tuned to answer questions while browsing the web - [Gpt-oss-120b](https://everyllm.cpn.cleartrust.cc/m/gpt-oss-120b) (2025), United States: OpenAI's open-weight reasoning model that fits on one 80GB GPU - [Gpt-oss-20b](https://everyllm.cpn.cleartrust.cc/m/gpt-oss-20b) (2025), United States: OpenAI's smaller open-weight reasoning model for local and low-latency use - [Claude 1](https://everyllm.cpn.cleartrust.cc/m/claude-1), United States: Anthropic's first AI assistant model, offered in two versions via chat and API. - [Claude 2](https://everyllm.cpn.cleartrust.cc/m/claude-2) (2023), United States: Anthropic's 2023 assistant model with a 100K-token window and stronger coding. - [Claude 3 Opus](https://everyllm.cpn.cleartrust.cc/m/claude-3-opus), United States: Anthropic's largest Claude 3 model, aimed at complex reasoning and analysis - [Claude 3 Haiku](https://everyllm.cpn.cleartrust.cc/m/claude-3-haiku) (2024), United States: Anthropic's fastest and lowest-cost model in the Claude 3 family - [Claude 3.5 Haiku](https://everyllm.cpn.cleartrust.cc/m/claude-3-5-haiku) (2024), United States: Anthropic's fast, low-latency model for coding, tools and user-facing products - [Claude 3.7 Sonnet](https://everyllm.cpn.cleartrust.cc/m/claude-3-7-sonnet) (2025), United States: Anthropic's hybrid reasoning model with a controllable thinking budget - [Claude 4](https://everyllm.cpn.cleartrust.cc/m/claude-4) (2025), United States: Anthropic's fourth-generation coding and reasoning models, Opus 4 and Sonnet 4 - [Claude Opus 4.1](https://everyllm.cpn.cleartrust.cc/m/claude-opus-4-1) (2025), United States: Anthropic's upgraded Claude Opus 4 model for coding, agents and reasoning - [Claude Sonnet 4.5](https://everyllm.cpn.cleartrust.cc/m/claude-sonnet-4-5) (2025), United States: Anthropic's coding and agentic Claude model, released in 2025. - [Claude Haiku 4.5](https://everyllm.cpn.cleartrust.cc/m/claude-haiku-4-5) (2025), United States: Anthropic's small, fast Claude model for coding, agents and real-time work - [Claude Opus 4.5](https://everyllm.cpn.cleartrust.cc/m/claude-opus-4-5) (2025), United States: Anthropic's flagship Claude model for coding, agents and computer use - [Claude Sonnet 4.6](https://everyllm.cpn.cleartrust.cc/m/claude-sonnet-4-6), United States: Anthropic's Sonnet-tier model for coding, computer use and long-context work - [Claude Opus 5](https://everyllm.cpn.cleartrust.cc/m/claude-opus-5), United States: Anthropic's Claude Opus model for coding, agents and knowledge work - [Claude Opus 4.6](https://everyllm.cpn.cleartrust.cc/m/claude-opus-4-6) (2026), United States: Anthropic's largest Claude model, built for long agentic coding and knowledge work - [Gemini 1.5 Pro](https://everyllm.cpn.cleartrust.cc/m/gemini-1-5-pro) (2024), United States: Google DeepMind's mid-size multimodal model with up to a 1 million token context window - [Gemini 1.0](https://everyllm.cpn.cleartrust.cc/m/gemini-1-0) (2023), United States: Google's first Gemini models: multimodal AI in Ultra, Pro and Nano sizes - [Gemini 3 Pro](https://everyllm.cpn.cleartrust.cc/m/gemini-3-pro) (2025), United States: Google DeepMind's Gemini 3 Pro, a multimodal model with a 1 million-token context window. - [Gemini 3.1 Pro](https://everyllm.cpn.cleartrust.cc/m/gemini-3-1-pro) (2026), United States: Google DeepMind's multimodal reasoning model, the next step after Gemini 3 Pro - [Gemini 3.1 Flash-Lite](https://everyllm.cpn.cleartrust.cc/m/gemini-3-1-flash-lite) (2026), United States: A fast, low-cost multimodal reasoning model for high-volume tasks - [Gemini 3.5 Flash](https://everyllm.cpn.cleartrust.cc/m/gemini-3-5-flash) (2026), United States: A multimodal reasoning model from Google DeepMind with adjustable thinking levels - [Gemini Embedding](https://everyllm.cpn.cleartrust.cc/m/gemini-embedding) (2025), United States: Google's embedding model built from Gemini for text, code and many languages - [Gemma 2](https://everyllm.cpn.cleartrust.cc/m/gemma-2) (2024), United States: Open lightweight language models from Google DeepMind, 2B to 27B parameters - [Gemma](https://everyllm.cpn.cleartrust.cc/m/gemma-1) (2024), United States: Small open-weight language models from Google, in 2B and 7B sizes - [Gemma 3](https://everyllm.cpn.cleartrust.cc/m/gemma-3) (2025), United States: Lightweight multimodal open models from 1B to 27B parameters - [Gemma 3n](https://everyllm.cpn.cleartrust.cc/m/gemma-3n) (2025), United States: Mobile-first open model family for on-device text, image and audio work - [Gemma 4](https://everyllm.cpn.cleartrust.cc/m/gemma-4), United States: Google DeepMind's open multimodal models in five sizes, from phones to servers - [PaliGemma](https://everyllm.cpn.cleartrust.cc/m/paligemma) (2024): A 3B open vision-language model built on SigLIP and Gemma-2B - [RecurrentGemma](https://everyllm.cpn.cleartrust.cc/m/recurrentgemma) (2024), United States: Open language models using Google's Griffin recurrent-plus-attention architecture - [MedGemma](https://everyllm.cpn.cleartrust.cc/m/medgemma) (2025), United States: Medical vision-language models built on Gemma 3 for images and text - [Med-PaLM](https://everyllm.cpn.cleartrust.cc/m/med-palm) (2022), United States: A medical-domain language model aligned from Flan-PaLM using instruction prompt tuning. - [MT5](https://everyllm.cpn.cleartrust.cc/m/mt5) (2020), United States: A multilingual text-to-text transformer pretrained on 101 languages - [Switch Transformer](https://everyllm.cpn.cleartrust.cc/m/switch-transformer) (2021): A sparse mixture-of-experts language model scaling to a trillion parameters. - [Med-PaLM 2](https://everyllm.cpn.cleartrust.cc/m/med-palm-2) (2023), United States: Google's medical question-answering model built on PaLM 2 - [GLaM](https://everyllm.cpn.cleartrust.cc/m/glam) (2021), United States: A trillion-parameter mixture-of-experts language model from Google Research - [Minerva](https://everyllm.cpn.cleartrust.cc/m/minerva-google) (2022), United States: A language model trained further on technical content to solve maths and science problems - [ALBERT](https://everyllm.cpn.cleartrust.cc/m/albert) (2019), United States: A slimmed-down BERT encoder that shares weights across its layers - [MuRIL](https://everyllm.cpn.cleartrust.cc/m/muril) (2021), India: A BERT-style encoder pre-trained on 17 Indian languages and their transliterations - [Flamingo](https://everyllm.cpn.cleartrust.cc/m/flamingo) (2022), United Kingdom: A visual language model family that learns new image and video tasks from a few examples. - [Llama 2](https://everyllm.cpn.cleartrust.cc/m/llama-2) (2023), United States: Meta's open family of pretrained and chat-tuned language models, 7B to 70B - [Gopher](https://everyllm.cpn.cleartrust.cc/m/gopher) (2021), United Kingdom: A 280 billion parameter Transformer language model from DeepMind, studied for how scale changes ability. - [Llama 3.1 405B](https://everyllm.cpn.cleartrust.cc/m/llama-3-1-405b) (2024), United States: Meta's largest openly available language model, released in 2024 - [Llama 3.2](https://everyllm.cpn.cleartrust.cc/m/llama-3-2) (2024), United States: Small text models and vision models from Meta for edge devices and image tasks - [Phi-4](https://everyllm.cpn.cleartrust.cc/m/phi-4) (2024), United States: A 14B Microsoft Research model trained heavily on synthetic data for reasoning. - [Code Llama](https://everyllm.cpn.cleartrust.cc/m/code-llama) (2023), United States: Meta's code-specialised family of Llama 2 models in four sizes - [Llama Guard](https://everyllm.cpn.cleartrust.cc/m/llama-guard) (2023): A Llama 2 based model that classifies prompts and replies for safety risks - [BlenderBot 3](https://everyllm.cpn.cleartrust.cc/m/blenderbot-3) (2022): A 175B open-domain chatbot with internet access and long-term memory - [RoBERTa](https://everyllm.cpn.cleartrust.cc/m/roberta) (2019), United States: A robustly optimised BERT pretraining study with released models and code - [BART](https://everyllm.cpn.cleartrust.cc/m/bart) (2019), United States: A denoising encoder-decoder transformer for generation and comprehension tasks - [XLM-RoBERTa](https://everyllm.cpn.cleartrust.cc/m/xlm-roberta) (2019), United States: A multilingual masked language model covering 100 languages, built on RoBERTa. - [NLLB-200](https://everyllm.cpn.cleartrust.cc/m/nllb-200) (2022), United States: A research translation model covering 200 languages, built for low-resource ones. - [Chameleon](https://everyllm.cpn.cleartrust.cc/m/chameleon) (2024): Early-fusion models that read and write images and text in one stream - [Phi-1](https://everyllm.cpn.cleartrust.cc/m/phi-1) (2023), United States: A small Microsoft model specialised for basic Python coding - [DeepSeek-V2](https://everyllm.cpn.cleartrust.cc/m/deepseek-v2) (2024), China: Efficient mixture-of-experts language model with 236 billion parameters - [ALLaM](https://everyllm.cpn.cleartrust.cc/m/allam) (2024), Saudi Arabia: Bilingual Arabic and English language model series - [OLMoE](https://everyllm.cpn.cleartrust.cc/m/olmoe) (2024), United States: Open mixture-of-experts language model with 1 billion active parameters - [CyberAgentLM3](https://everyllm.cpn.cleartrust.cc/m/cyberagentlm3), Japan: A 22-billion parameter Japanese and English chat model by CyberAgent. - [Gemini 2.0 Flash](https://everyllm.cpn.cleartrust.cc/m/gemini-2-0-flash) (2024), United States: Google DeepMind's fast multimodal workhorse model for agentic apps - [Step-5-Preview](https://everyllm.cpn.cleartrust.cc/m/step-5-preview) (2026): Agentic model with 600B parameters and 1M token context - [HyperCLOVA](https://everyllm.cpn.cleartrust.cc/m/hyperclova) (2021), South Korea: Korean language model with 82 billion parameters - [K-EXAONE](https://everyllm.cpn.cleartrust.cc/m/k-exaone), South Korea: LG AI Research multilingual Mixture-of-Experts model - [Granite Code](https://everyllm.cpn.cleartrust.cc/m/granite-code) (2024), United States: IBM's open code models from 3B to 34B parameters for coding tasks - [InternLM2](https://everyllm.cpn.cleartrust.cc/m/internlm2) (2024), China: Open-source 20 billion parameter language model with long context support - [Meltemi](https://everyllm.cpn.cleartrust.cc/m/meltemi) (2024), Greece: Open large language model for Greek - [Voxtral](https://everyllm.cpn.cleartrust.cc/m/voxtral) (2025), France: Open audio chat models that transcribe, answer questions and summarise speech - [LLaVA](https://everyllm.cpn.cleartrust.cc/m/llava) (2023), United States: Open-source multimodal chatbot connecting vision encoders to language models. - [PaLM-E](https://everyllm.cpn.cleartrust.cc/m/palm-e) (2023): An embodied multimodal language model that grounds words in robot sensor input - [Mi:dm 2.0](https://everyllm.cpn.cleartrust.cc/m/mi-dm-2-0) (2025), South Korea: Korea-centric language model by KT - [MiniMax-M2.7](https://everyllm.cpn.cleartrust.cc/m/minimax-m2-7), China: Open-weight model for coding and agent tasks - [RWKV-4](https://everyllm.cpn.cleartrust.cc/m/rwkv-4) (2023): A dense recurrent model with transformer-level performance - [GPT-6](https://everyllm.cpn.cleartrust.cc/m/gpt-6-astra) (2026), United States: OpenAI family of large language models released in late 2026. - [OpenAssistant](https://everyllm.cpn.cleartrust.cc/m/openassistant) (2023), Germany: An open source chatbot built by volunteers and LAION. - [Devin](https://everyllm.cpn.cleartrust.cc/m/devin-ai), United States: Autonomous software engineering agent by Cognition Labs - [DeepSeek-R1](https://everyllm.cpn.cleartrust.cc/m/deepseek-r1) (2024), China: Open-weight reasoning model from China - [ChatGPT](https://everyllm.cpn.cleartrust.cc/m/chatgpt-agent) (2022), United States: Generative AI chatbot by OpenAI - [Claude Sonnet 4](https://everyllm.cpn.cleartrust.cc/m/claude-sonnet-4) (2025), United States: Mid-tier reasoning model from Anthropic released in May 2025 - [Jamba](https://everyllm.cpn.cleartrust.cc/m/jamba) (2024), Israel: Hybrid open-weights language model with long context - [Claude Opus 4](https://everyllm.cpn.cleartrust.cc/m/claude-opus-4) (2025), United States: High-capability reasoning model from Anthropic released in May 2025.