News
ResearchOpenAI pens AI safety agreement with four firmsReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierOpenAI鈥檚 New Hacking Model Found Dozens Of Holes In Apple TechFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIResearchOpenAI pens AI safety agreement with four firmsReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierOpenAI鈥檚 New Hacking Model Found Dozens Of Holes In Apple TechFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAI
EveryLLM

Dense decoder transformers

The GPT design: one stack of transformer layers predicting the next token. 308 models on EveryLLM so far.

The dense decoder transformer is the design behind GPT and most chatbots. It is a single stack of identical transformer layers that reads the tokens so far and predicts the next one, with a mask that stops each position from looking ahead. Dense means every parameter is used for every token, unlike in a mixture of experts. GPT-3, PaLM, Llama 3, Chinchilla, Gemma, Phi and OLMo are dense decoders.

Inside each layer

Each layer has two parts. Self-attention lets every token weigh every earlier token and gather what it needs; a feed-forward network then transforms each token's vector on its own. The transformer was introduced by Google researchers in the June 2017 paper Attention Is All You Need, for translation, and OpenAI's GPT-1 (2018) kept only its decoder half. Llama 3.1 405B, released in July 2024, stacks 126 such layers.

Why it won

A decoder trains on any text without labels, because the next token is always the answer to predict, and it scales smoothly: scaling laws published by OpenAI in 2020 and DeepMind in 2022 showed the error falling predictably with more parameters, data and compute. Its weak point is cost. Attention grows with the square of the context length, and every parameter must be read for every token, which is why mixtures of experts became popular.

Famous dense decoder models

Still to be written: OLMo (language model).

Further reading

XLNet WorldXLNet Permutation-based autoregressive transformer model created by Google AIResearch 路 Open weights 路 2019 1.58-bit large language model World1.58-bit large language model A model whose weights hold only three values: -1, 0 and +1.Research 路 Open weights 路 2024 Minerva 馃嚠馃嚬 ItalyMinerva An Italian-first large language model trained from scratch at Sapienza University of Rome.General 路 Open weights 路 2024 fnGLM 馃嚚馃嚦 ChinaGLM Open-weight Chinese model series used for chat and codingGeneral 路 Open weights 路 2021 imgwavvidQwen 馃嚚馃嚦 ChinaQwen A family of large and small language models developed by Alibaba Cloud.General 路 Open weights 路 2023 Velvet AI 馃嚠馃嚬 ItalyVelvet AI Italian multilingual open-weight LLMs built for energy efficiencyGeneral 路 Open weights 路 2025 imgfnPhi 馃嚭馃嚫 United StatesPhi Microsoft's family of small, open-weight language models that run on everyday devices.Small 路 Open weights 路 2023 imgwavvidGemma 馃嚭馃嚫 United StatesGemma Google DeepMind's open family of Gemini-derived language modelsGeneral 路 Open weights 路 2024 GPT-1 WorldGPT-1 OpenAI's first GPT model, a 117M-parameter decoder-only transformer from 2018.Research 路 Open weights 路 2018 Jais 馃嚘馃嚜 United Arab EmiratesJais An open Arabic and English large language model from the UAEGeneral 路 Open weights 路 2023 GPT-J WorldGPT-J A 6-billion-parameter open English text generator from EleutherAI, released in 2021.General 路 Open weights 路 2021 IBM Granite 馃嚭馃嚫 United StatesIBM Granite IBM's decoder-only foundation models, from 3 to 34 billion parametersGeneral 路 Open weights 路 2023 GPT-4chan WorldGPT-4chan A GPT-J model fine-tuned on posts from 4chan's /pol/ board.General 路 Open weights 路 2022 imgHuawei PanGu 馃嚚馃嚦 ChinaHuawei PanGu Huawei's multimodal large model family for industry, from 1B to trillion-scaleMultimodal 路 Open weights 路 2023 GPT-2 馃嚭馃嚫 United StatesGPT-2 OpenAI's 2019 text model that predicts the next word in a sequenceGeneral 路 Open weights 路 2019 Apertus 馃嚚馃嚟 SwitzerlandApertus A Swiss public large language model released in 8B and 70B sizesGeneral 路 Open weights 路 2025 Llama 2 馃嚭馃嚫 United StatesLlama 2 Meta's open family of pretrained and chat-tuned language models, 7B to 70BGeneral 路 Open weights 路 2023 SOLAR 10.7B 馃嚢馃嚪 South KoreaSOLAR 10.7B An 11-billion parameter language model using depth up-scaling.General 路 Open weights 路 2023 imgwavvidGemma 4 馃嚭馃嚫 United StatesGemma 4 Google DeepMind's open multimodal models in five sizes, from phones to serversGeneral 路 Open weights imgvidfnGemma 3 馃嚭馃嚫 United StatesGemma 3 Lightweight multimodal open models from 1B to 27B parametersGeneral 路 Open weights 路 2025 imgMistral Medium 3.5 馃嚝馃嚪 FranceMistral Medium 3.5 A 128B multimodal model with a 256k window for chat, reasoning and codingGeneral 路 Open weights Project Indus 馃嚠馃嚦 IndiaProject Indus Open-source language model for Hindi and dialectsGeneral 路 Open weights Gemma 2 馃嚭馃嚫 United StatesGemma 2 Open lightweight language models from Google DeepMind, 2B to 27B parametersGeneral 路 Open weights 路 2024 Falcon 180B 馃嚘馃嚜 United Arab EmiratesFalcon 180B Open 180 billion parameter language model from TIIGeneral 路 Open weights 路 2023 Swallow 馃嚡馃嚨 JapanSwallow Japanese-enhanced Llama models by Tokyo TechGeneral 路 Open weights 路 2024 SEA-LION 馃嚫馃嚞 SingaporeSEA-LION Multilingual language models for Southeast AsiaGeneral 路 Open weights 路 2025 Falcon-7B 馃嚘馃嚜 United Arab EmiratesFalcon-7B Open 7-billion parameter decoder model by TIIGeneral 路 Open weights 路 2023 Sarashina2 馃嚡馃嚨 JapanSarashina2 Open-weight Japanese and English language models by SB IntuitionsGeneral 路 Open weights RakutenAI-7B 馃嚡馃嚨 JapanRakutenAI-7B Japanese-focused open language model based on MistralGeneral 路 Open weights 路 2024 PLLuM 馃嚨馃嚤 PolandPLLuM Polish-focused large language model familyGeneral 路 Open weights 路 2024 imgYi-34B 馃嚚馃嚦 ChinaYi-34B A 34-billion parameter bilingual language model by 01.AIGeneral 路 Open weights 路 2023 Mistral 7B 馃嚝馃嚪 FranceMistral 7B A 7.3 billion parameter open language model from Mistral AIGeneral 路 Open weights 路 2023 Sarvam-M 馃嚠馃嚦 IndiaSarvam-M Hybrid reasoning model for Indian languages and EnglishGeneral 路 Open weights 路 2025 Falcon 3 馃嚘馃嚜 United Arab EmiratesFalcon 3 Open foundation models from 1B to 10B parameters by TIIGeneral 路 Open weights 路 2024 DialoGPT 馃嚭馃嚫 United StatesDialoGPT A GPT-style model trained on Reddit threads to write chat repliesGeneral 路 Open weights 路 2019 Falcon2-11B 馃嚘馃嚜 United Arab EmiratesFalcon2-11B An 11 billion parameter causal decoder model by TII.General 路 Open weights 路 2024 Phi-2 馃嚭馃嚫 United StatesPhi-2 A 2.7 billion parameter Microsoft research model for QA, chat and codeResearch 路 Open weights Phi-4-reasoning 馃嚭馃嚫 United StatesPhi-4-reasoning A 14B open-weight reasoning model fine-tuned from Phi-4Reasoning 路 Open weights 路 2025 imgPhi-3 馃嚭馃嚫 United StatesPhi-3 Small Microsoft language models built to run well on phones and laptops.Small 路 Open weights 路 2024 INTELLECT-1 WorldINTELLECT-1 A 10 billion parameter model trained by a global community.General 路 Open weights 路 2024 fnBlenderBot 3 WorldBlenderBot 3 A 175B open-domain chatbot with internet access and long-term memoryGeneral 路 Open weights 路 2022 Mixtral 8x7B 馃嚝馃嚪 FranceMixtral 8x7B Sparse mixture-of-experts language model with open weightsGeneral 路 Open weights 路 2023 Pythia 馃嚭馃嚫 United StatesPythia A suite of 16 models for interpretability research.Research 路 Open weights 路 2023 GLM-130B 馃嚚馃嚦 ChinaGLM-130B Open bilingual pre-trained language model with 130 billion parametersGeneral 路 Open weights 路 2022 imgfnGLM-4 馃嚚馃嚦 ChinaGLM-4 Multilingual large language model series by Zhipu AIGeneral 路 Open weights 路 2024 OpenHathi 馃嚠馃嚦 IndiaOpenHathi A 7 billion parameter language model for Hindi and English.General 路 Open weights StarCoder2 WorldStarCoder2 Open code generation model by BigCodeCode 路 Open weights 路 2024 fnMagistral 馃嚝馃嚪 FranceMagistral Mistral AI's first reasoning model, in open and enterprise versionsReasoning 路 Open weights 路 2025

The weekly digest

Every Monday: the model of the day, new pages on EveryLLM, and the conferences and anniversaries of the week ahead.