Latest / Frontier models and safety

Model releaseFrontierChinaConfirmed

MiniMax releases M3.1 Flash, an open-weights model for agents, coding and cybersecurity

See how this story connects

MiniMax released M3.1 Flash, a multimodal mixture-of-experts model with about 428 billion parameters, 23 billion active per token, and a context window of up to 1 million tokens, with weights on Hugging Face under the MiniMax community licence. The model card reports 73.7% on DeepSWE v1.1, 61.7 on OSWorld V2 and 82.2 on the CyberGym cybersecurity benchmark. MiniMax says inference is nearly three times faster than before and usage costs 70% less, and that an early version ran anonymously as Space Bunny and processed more than 200 trillion tokens.

Why it matters

A Chinese lab has put weights for a model it markets on cybersecurity performance in anyone’s hands, as US labs gate their most cyber-capable models.

When
17 Oct 2026
ID
WTA-N-2026-0401
Where
China
Who
MiniMax

Summary by wheresthe.ai, from MiniMax and the reporting credited above. Checked against the source on 24 Oct 2026.

Source copy: download the page as saved on 24 Oct 2026 (PDF). Also held by the Internet Archive.

Next stop on this Storyline: What happened nextMoonshot AI releases Kimi K3.1 Pro with 1-million-token context window

Cite this story

WTA-N-2026-0401: “MiniMax releases M3.1 Flash, an open-weights model for agents, coding and cybersecurity.” wheresthe.ai, 17 Oct 2026. https://wheresthe.ai/d/cn-minimax-releases-m3-1-flash-open-weights-model-2026-10/ (source: MiniMax).

Spotted an error? Email [email protected].