News
ReleasesOVHcloud 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 improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior ReleaseReleasesOVHcloud 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 improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior Release
EveryLLM
MiniMax-Text-01

MiniMax-Text-01

A large language model with a four million token context window.

MiniMax-Text-01 is a large language model released by MiniMax in January 2025. It uses a hybrid architecture to handle contexts of up to four million tokens during inference. The model contains 456 billion parameters and matches the performance of leading commercial systems on standard benchmarks.

Share on WhatsApp

  • Handles up to 4 million tokens during inference.
  • Contains 456 billion total parameters.
  • Uses a hybrid of lightning and softmax attention.
  • Activates 45.9 billion parameters per token.
  • Released with open weights in January 2025.

What it is

MiniMax-Text-01 is a foundation model developed by the Chinese artificial intelligence company MiniMax. It was introduced in January 2025 as part of the MiniMax-01 series. The model is designed for general purpose tasks and features an exceptionally large context window. It contains 456 billion parameters in total, using a mixture of experts design where only 45.9 billion are active for any given token. This structure allows it to process vast amounts of text efficiently while maintaining high performance on academic tests.

How it came about

The model was announced in a technical report published on arXiv in January 2025. The paper lists over eighty authors from MiniMax, with names presented in alphabetical order. It was released alongside a vision-language variant called MiniMax-VL-01. The developers made the model weights publicly available through platforms like Hugging Face at the time of launch, encouraging community use and deployment.

Inside the model

The architecture combines softmax attention with lightning attention, placing a softmax layer after every seven lightning layers. It uses rotary position embeddings with a high base frequency to manage long sequences. The model employs thirty two experts within its mixture of experts framework. Training utilized advanced parallel strategies such as linear attention sequence parallelism plus and expert tensor parallel. These methods allowed the team to train on one million tokens and extend inference capacity to four million tokens.

What it can do

On standard benchmarks, the model performs comparably to top tier systems like GPT-4o and Claude-3.5-Sonnet. It scored 88.5 on MMLU and 75.7 on MMLU-Pro. In long context tests, it maintained strong retrieval accuracy, scoring 0.910 on the one million token needle in a haystack test. It also achieved high scores on LongBench v2, outperforming several competitors in overall and hard categories. The model supports function calling with structured JSON output.

How to use it

The weights are open and available for download. Developers can load the model using the Transformers library with recommended int8 quantization. For production environments, the creators suggest using vLLM for serving due to its memory management and throughput capabilities. The model integrates with standard tools and supports OpenAI compatible function calling definitions.

Questions people ask about MiniMax-Text-01

How many parameters does MiniMax-Text-01 have?

The model has 456 billion total parameters. However, it uses a mixture of experts architecture, so only 45.9 billion parameters are activated for each token processed.

What is the maximum context length?

The model was trained on a context length of one million tokens. During inference, it can handle inputs of up to four million tokens, allowing it to process entire books or large codebases.

Is the model open source?

Yes, the model weights are publicly released. Developers can download them from Hugging Face and run them locally using libraries like Transformers or serve them via vLLM.

Good reading and sources

    Where this page comes from

    There is no English Wikipedia article on MiniMax-Text-01 yet, so the facts on this page come from the sources listed above: huggingface.co/MiniMaxAI/MiniMax-Text-01, arxiv.org/abs/2501.08313. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryLLM wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.

    It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

    The weekly digest

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