MiniMax-M3 is the next generation of the MiniMax series, building on the agent harness, software engineering, and professional-work foundations established by MiniMax-M2.7. The model is not yet released — this repository exists so the community can share what they need next.
Before M3 lands, we are listening. If you are using MiniMax-M2.7 (via the API, Agent, or locally) and have something to say about it, please tell us — every report directly shapes M3.
We are especially interested in:
- 🐛 Bugs and regressions — anything that broke, hallucinated, or behaved unexpectedly in M2.7.
- 💡 Capability requests — what M2.7 still can't do well for your workload (agent harnesses, SWE, professional work, entertainment, multilingual, long context, tool use, …).
- 📊 Benchmark gaps — public or internal evals where you would like to see M3 improve.
- 🧰 Deployment pain points — issues with SGLang, vLLM, Transformers, ModelScope, NIM, or the API.
- 🧠 Agent / skill feedback — anything you observed while building Agent Teams, Skills, or dynamic tool search on top of M2.7.
| Channel | Use for |
|---|---|
| 📮 Open an Issue | Bugs, capability requests, M2.7 → M3 comparisons. Pick a template. |
| Chinese-speaking community discussion. | |
| 🧩 Discord | English-speaking community discussion. |
| ✉️ model@minimax.io | Private feedback, partnership, or evaluation requests. |
If you are reporting a bug from M2.7, please include:
- Which inference path you used (MiniMax API / Agent / SGLang / vLLM / Transformers / NIM / ModelScope).
- Inference parameters (
temperature,top_p,top_k, system prompt). - A minimal reproduction — prompt, expected output, actual output.
While M3 is in development, M2.7 remains our latest released model:
- MiniMax Agent: https://agent.minimax.io/
- MiniMax API: https://platform.minimax.io/
- Token Plan: https://platform.minimax.io/subscribe/token-plan
- Weights & deployment guides: MiniMax-M2.7 (SGLang / vLLM / Transformers / ModelScope / NVIDIA NIM)
- Model card: https://huggingface.co/MiniMaxAI/MiniMax-M2.7
Recommended inference parameters for M2.7: temperature=1.0, top_p=0.95, top_k=40.
Watch this repository for the M3 announcement, release notes, weights, and deployment guides.
Contact us at model@minimax.io.