Reasoning

MiniMax M2 Non-Commercial License (modified MIT base)Open weightsUpdated June 2026Frontier 2026

MiniMax M2.5

228.7B-parameter sparse mixture-of-experts model (8 experts active per token) in MiniMax's M2 line, released between M2.1 and M2.7, with a 196,608-token context window.

MiniMax · MiniMax

Editorial review

Reviewed byOpenSourcesAI EditorialLast updatedJune 2026SourcesMiniMaxAI/MiniMax-M2.5 on Hugging Face

Model checkpoints, context windows, provider support, local runtime compatibility, and license terms can change quickly. Verify the exact model card before production or commercial use.

Best for

Builders tracking the M2 line's iteration for coding and agent workflows; M2.7 supersedes it with a larger context window if a newer release is available.

Who should use it

  • Builders tracking the M2 line's iteration for coding and agent workflows; M2.7 supersedes it with a larger context window if a newer release is available.
  • Teams with access to hosted inference or server-class deployment paths.
  • Developers evaluating coding assistant, repo-editing, and code review workflows.
  • Teams testing tool-use, agentic planning, and multi-step workflow behavior.

Common workflows

  • Coding, reasoning, tool use, office workflows, agents
  • reasoning workflows
  • coding workflows
  • tool-use workflows
  • agents workflows

Deployment and hardware notes

228.7B total parameters, 8 experts active per token. No GGUF in the official repository -- server-class multi-GPU or hosted inference only.

License and usage notes

MiniMax M2 Non-Commercial License (modified MIT base). Open weights. Verify the exact model card and license terms for the checkpoint or hosted provider you use.

Strengths

  • Open weights model option for MiniMax workflows.
  • Builders tracking the M2 line's iteration for coding and agent workflows; M2.7 supersedes it with a larger context window if a newer release is available.
  • Tracked as Frontier 2026 in the OpenSourcesAI model directory.

Limitations

  • Same non-commercial licensing terms as the rest of the M2 line: prior written authorization required for commercial use. No official or community GGUF -- server-class or hosted inference only.
  • 228.7B total parameters, 8 experts active per token. No GGUF in the official repository -- server-class multi-GPU or hosted inference only.
  • Context window and limits: 196,608 tokens, confirmed from the model's published config.
  • Verify the exact model card, provider docs, license, and serving support before production use.

Frontier-model verification note

This page is written to stay accurate as of the latest available 2026 public model information. Availability, licenses, context windows, API support, pricing, benchmark standing, and local-serving support can change quickly. Verify the official model card, provider docs, and license before using this model in production or commercial workflows.

Sources to verify

MiniMaxAI/MiniMax-M2.5 on Hugging Face

Related resources

Continue with model source notes, local tools, and implementation guides related to this model.

MiniMax family hubModel selection guideFrontier vs open-weight contextLocal coding assistant stackOpen-source AI coding assistants
HardwareServer-classRuntimeHosted providers, vLLM where supportedContext196,608 tokens, confirmed from the model's published configLast updated2026
MiniMaxAI/MiniMax-M2.5 on Hugging Face →

Model ecosystem connections

Use these next-step links to move from this profile into related tools, comparisons, guides, stacks, and curated shortlists.

Recommended runtimes and tools

vLLM →

Setup and deployment

RunPod — managed GPU cloud →Lambda Labs — A100 / H100 on demand →Vast.ai — crowdsourced GPU marketplace →

Related model pages

MiniMax family hub →MiniMax M2.7 →MiniMax M3 →MiniMax M1 →DeepSeek R1 →DeepSeek R1 Distill Llama 70B →DeepSeek R1 Distill Llama 8B →

Guides, stacks, and comparisons

Model selection guide →Frontier vs open-weight context →Local coding assistant stack →Open-source AI coding assistants →Cursor vs Windsurf vs Cline →What is MCP? →

Related models

MiniMax M2.7Builders comparing current agent-ready open-weight models for coding and productivity workflows, for non-commercial or internally-authorized use.MiniMax M3Teams evaluating frontier-scale open-weight reasoning/coding models with genuine multimodal (image) input and a ~1M-token context window, willing to work within a named commercial license rather than a standard OSI one.MiniMax M1Builders who specifically want a permissively-licensed (Apache 2.0) MiniMax release for commercial use, at the cost of a larger, older architecture than the M2.x/M3 lines.DeepSeek R1Reasoning experiments, coding workflows, and evaluating open reasoning models against closed alternatives.DeepSeek R1 Distill Llama 70BReasoning experiments, local-friendly evaluation, and distilled model comparisons.DeepSeek R1 Distill Llama 8BReasoning experiments, local-friendly evaluation, and distilled model comparisons.

Ready to run this model locally?

Find a compatible interface in our Local AI Tools directory →