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
fnMiniMax-M2

MiniMax-M2

Open-source MoE model for coding and agents

MiniMax-M2 is an open-source mixture of experts language model released by MiniMax AI. It features 230 billion total parameters with 10 billion active per token. The model targets efficient coding workflows and agentic tool use, offering competitive benchmark scores while maintaining low latency for deployment.

Share on WhatsApp

  • 230 billion total parameters with 10 billion active
  • Top composite score among open-source models per AA
  • Supports 128k token context window
  • Optimized for coding and agentic tool use
  • Weights available openly on Hugging Face

What it is

MiniMax-M2 is a large language model designed for coding and autonomous agent tasks. Developed by MiniMax AI, it uses a mixture of experts architecture to balance size and speed. The model contains 230 billion parameters in total but activates only 10 billion during inference. This design aims to provide high intelligence for software development and tool usage while keeping computational costs manageable for developers.

How it came about

MiniMax AI released MiniMax-M2 as an open-source model. The company made the weights available on Hugging Face for local deployment. Alongside the model release, they launched an API on their open platform and a product called MiniMax Code. The API and product were offered free for a limited time to encourage adoption among developers and researchers.

Inside the model

The model employs a mixture of experts structure, a design where only a subset of neural network parameters processes each input. This results in 10 billion active parameters per token out of 230 billion total. The architecture supports a context window of 128,000 tokens. This efficient activation size allows for faster feedback loops in coding tasks and reduces memory requirements for running multiple concurrent agent instances.

What it can do

Benchmarks from Artificial Analysis place its composite intelligence score at the top among open-source models. It performs strongly in coding tests like SWE-bench Verified, scoring 69.4, and Terminal-Bench, with a score of 46.3. The model handles multi-file edits and complex toolchains involving web browsing and shell commands. It also shows competence in mathematics and science tasks, competing with larger proprietary models in specific agentic evaluations.

How to use it

Developers can access MiniMax-M2 through three main channels. The model weights are open source and downloadable from Hugging Face for local use. An API is available via the MiniMax Open Platform. Additionally, the company offers MiniMax Code, a product built on the model. Both the API and the product were free for a limited period following the release.

Questions people ask about MiniMax-M2

What is the parameter size of MiniMax-M2?

The model has 230 billion total parameters. However, it uses a mixture of experts architecture that activates only 10 billion parameters for each token processed. This design improves efficiency and reduces latency.

Is MiniMax-M2 available for free?

Yes, the model weights are open source and can be downloaded from Hugging Face. The API and the MiniMax Code product were also offered free for a limited time after the model's release.

What tasks is MiniMax-M2 best suited for?

It is engineered for coding workflows and agentic tasks. This includes multi-file code editing, executing terminal commands, and using tools like web browsers. It maintains strong general intelligence across math and science benchmarks.

Good reading and sources

    Where this page comes from

    There is no English Wikipedia article on MiniMax-M2 yet, so the facts on this page come from the sources listed above: huggingface.co/MiniMaxAI/MiniMax-M2. 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.