MiniMax M3
MiniMaxMiniMax M3 launched 1 Jun 2026: API-first at launch with MiniMax Sparse Attention, 1M-token context, native multimodal. Open weights have since been released on Hugging Face (ungated, ~131k downloads/month) under a custom minimax-community license; training data and pipeline remain closed.
MiniMax M3 launched 1 Jun 2026: API-first at launch with MiniMax Sparse Attention, 1M-token context, native multimodal. Open weights have since been released on Hugging Face (ungated, ~131k downloads/month) under a custom minimax-community license; training data and pipeline remain closed. Consolidated on 2026-07-29 from minimax-m2-5; openness follows minimax-m3, the release that currently governs. See sources/slug_aliases.yaml.
Openness
3 high confidence- weights
- open(on HF
- data
- closed
- code
- partial(inference only
- license
- minimax-community(custom, see HF LICENSE)
Weights distributed under the custom minimax-community license with a closed corpus, so 3. M3 governs as the current release. Capability 5 is carried from M2.5 because M3 records no capability score at all, so the tier's capability describes the older release and should be re-read when M3 is benchmarked - it is the one axis here not sourced to the governing release.
- https://huggingface.co/api/models/MiniMaxAI/MiniMax-M3 recorded 2026-06-23
Public ungated weights repo (safetensors/MoE), license=minimax-community, created 2026-06-02
- https://venturebeat.com/technology/minimax-m3-debuts-eclipsing-gpt-5-5-and-gemini-3-1-pro-on-key-benchmark-performance-for-just-5-10-of-the-cost recorded 2026-06-04
flagship phase-C verification source
Adoption
3 high confidenceHF model card reports ~693,331 downloads in the last month (100K-1M band); additionally served via OpenRouter and free in coding tools (Kilo Code). Headline figure is the HF monthly download count.
- https://huggingface.co/MiniMaxAI/MiniMax-M2.5 recorded 2026-06-04
~693,331 downloads last month on the official HF model page
Capability
5 medium confidenceFrontier-tier among open-weights agentic chat models (SWE-bench Verified 80.2% with strong reasoning/browse scores), at very low serving cost. Figures are MiniMax-reported on the HF card; not independently audited but consistent with peer open models (Kimi K2.6, DeepSeek-V4).
- https://huggingface.co/MiniMaxAI/MiniMax-M2.5 recorded 2026-06-04
MiniMax-reported SWE-bench Verified 80.2%, GPQA-D 85.2%, AIME25 86.3% on the official HF card
Unchanged since 2026-07-30 (last edited, not re-checked)