MiniMax-M2.7
MiniMax·
LLMsopen-weightMulti-GPU Self-Host
Overview
Open-weight agentic MoE: 230B total / 10B active, 256 experts, ~200K context. Coding and tool-use focus.
Capabilities and innovations
230B total / 10B active (MoE)256 experts~200K context windowAgentic & coding focus
Benchmarks
- GPQA Diamond: independent evaluation pending
- Humanity’s Last Exam (no tools): independent evaluation pending
- MMLU-Pro: independent evaluation pending
- MMMU-Pro: independent evaluation pending
Architecture and hardware
- Parameters
- 230B total, 10B active per token (MoE)
- Estimated VRAM at Q4
- ~133 GB, Multi-GPU Self-Host class
- Quantization formats
- FP8, GGUF
- Recommended runtime
- vLLM
- License
- Custom (MiniMax Model License)
Links
Data curated by AI Model Timeline. See the methodology for admission criteria, benchmark eras and source priority.