MiniMax open-sources M3.1 Flash: 428B total params, 23B active, built for agentic tasks
VoidAsuka · x · 2026-10-23
MiniMax has released open weights for M3.1 Flash, a 428B-parameter MoE model with 23B activated per token, available on Hugging Face and ModelScope.
- Agentic performance: The company calls it a generational leap in agentic tasks, coding, and cybersecurity — 76.2 on BankerToolBench, 73.7% on DeepSWE v1.1, 61.7 on OSWorld V2, 29.6 on Agents' Last Exam, and 82.2 on CyberGym.
- Efficiency: Uses MiniMax Sparse Attention and DSpark speculative decoding for faster inference.
- Ecosystem: Ships with an API, M Plan subscription, and MCode desktop app.
The model card also reveals its tool-calling template: tool arguments render as XML, with None fields omitted entirely (never emitting literal "null"), plus upstream cleaning as a safety net.
Related event: MiniMax Open-Sources M3.1-Flash: 428B-Parameter MoE with 1M Context(9 posts)→
More from Models
- 345K trajectories reveal LLM agent personalities: Claude most planful, GPT-5.5 least energetic — mdredze · 2026-10-23
- DeepSeek V4.1 Coding Benchmark Labels Look Wrong: 47 vs 46 Score Confusion — teortaxesTex · 2026-10-23
- DeepSeek 4.1 Flash Hits ~80 tok/s Locally on DGX Station via LM Studio — MatthewBerman · 2026-10-23
- MiniMax M3.1 Flash launches day-zero on Modular Cloud, tops open models in cowork automation — MiniMax_AI · 2026-10-23
- Neolab ecosystem ships flurry of updates: $1B Series A, new coding model Flint — FinanceYF5 · 2026-10-23
- Recursive Flint benchmarks: matches Qwen3.8-27B on agentic tasks, weak on long context — ChengleiSi · 2026-10-23