MiniMax-M2.1

2.1

Open-weight MoE language model (~229B parameters) by MiniMax, optimized for agentic coding, tool use and long-horizon planning.

✓ Active✓ Public access⚖ Open sourceLLMReasoning modelTool-using model

Context window

192K

tokens

Parameters

229B

parameters

Release date

20 December 2025

🏢MiniMaxProducer

Access:APIDownloadHostedDeployment:💻 Local☁ Cloud

Overview

MiniMax-M2.1 is an open-source Mixture-of-Experts (MoE) language model developed by the Chinese company MiniMax. It has roughly 229 billion total parameters, with a sparse MoE design of 256 experts of which 8 are activated per token.

The model is optimized for agentic tasks: coding, tool use, instruction following and long-horizon planning. It supports a context window of up to 196,608 tokens (about 192K), uses grouped-query attention (GQA), RoPE positional encoding and multi-token prediction (MTP), and its weights are distributed in FP8 format.

The weights are publicly available on Hugging Face under a Modified-MIT license, with API access provided through the MiniMax Open Platform. The MiniMax Agent product is built on this model. According to the model card, M2.1 improves over M2 on software-engineering benchmarks, including SWE-bench Verified (74.0).

Classification

LLMReasoning modelTool-using model

Access & deployment

APIDownloadHosted

LocalCloud

Weights: Open source

Key parameters

📏 Context: 192K

🧩 Parameters: 229B

✓ Tools · ✓ Fine-tuning

📥 Input: text

Technical specification

Context window

192K

tokens

Parameters

229B

parameters

License

Modified MIT

Hardware requirements

Distributed as FP8 safetensors (~229B total MoE parameters); intended for multi-GPU deployment (e.g. vLLM / SGLang).

Features:✓ Tool use✓ Fine-tuning

Modalities

⬇ Input

text

⬆ Output

textcode

Capabilities and applications

Native model capabilities

Coding

Generating, analysing and modifying code in many programming languages. Covers writing functions, debugging, refactoring, code review, and creating tests. Measured by benchmarks such as HumanEval and SWE-bench.

Category: coding

Agentic coding

Multi-hour, multi-step programming tasks performed autonomously by the model: cloning a repository, running tests, iterating on fixes, integrating with CLI tools. Characteristic of Codex variants (GPT-5.1-Codex-Mini, Codex-Max).

Category: coding

Agentic capability

The model's ability to autonomously plan and execute multi-step tasks by sequentially using tools, maintaining context, and adapting to intermediate results.

Category: planning

Tool use

The model's ability to call external functions, APIs and tools during a conversation: calculator, search engine, code editor, database. The model decides when and how to use a tool and interprets its result.

Category: planning

Function Calling

Category: planning

Instruction following

Precisely following instructions contained in the prompt: response format, length, style, constraints (e.g. 'reply in six words'). GPT-5.1 significantly improved this capability compared to GPT-5.

Category: language

Planning

Forming and executing action plans for complex tasks.

Category: planning

Multi-step project execution

The ability to autonomously drive multi-hour, multi-step projects: decomposing the task, planning the sequence of actions, iteratively delivering results, and adjusting based on feedback. Key for enterprise and knowledge-work agents.

Category: planning

Long context

Support for large context windows — tens to hundreds of thousands (or millions) of input tokens. Enables analysis of entire codebases, long documents, and many parallel conversations without losing earlier information. GPT-5.1 supports 400,000 tokens.

Category: language

Multilingual

Competence in many natural languages (from a few to over a hundred): understanding, generation, translation, and code-switching within a single conversation. Frontier models support a wide range of languages with comparable quality.

Category: language

Reasoning

The model's ability to reason logically and solve complex problems.

Category: reasoning

Mathematical reasoning

The model's ability to solve mathematical tasks requiring multi-step reasoning — equations, proofs, combinatorics, geometry, calculus and competition-level problems.

Category: reasoning

MCP support

Native support for the Model Context Protocol - the model can integrate with external MCP servers, invoke their tools, and use their data sources without a dedicated wrapper.

Category: other

Web browsing

Ability of the model to autonomously search and browse web pages to retrieve up-to-date information.

Category: other

Benchmark results

11 benchmarks

SWE-bench

resolved · Verified

74.0%