← Back to live feed1 story
MiniMax published the underlying weights of its 428 billion parameter M3.1 Flash model on October 23 through Hugging Face and ModelScope, ending a stealth phase as the "Space Bunny" prototype that had processed more than 200 trillion tokens. The model supports a 1 million token context window and is specialized for multi-step automation, coding and cybersecurity, achieving inference speeds exceeding 1,000 tokens per second on Nvidia Blackwell and AMD Instinct GPUs using the SGLang Omni framework.
Sign in to suggest
Key sources
- SOURCEhuggingface.co
- SOURCE@minimax_ai“MiniMax M3.1 Flash weights are now public. 428B parameters total, 23B activated per token”x.com
- SOURCE@minimax_ai“M3.1 Flash achieves frontier-level intelligence across key tasks such as Cowork (Automation-Bench: 62%; ALE: 30%), Coding (DeepSWE: 74%) and Visual Reasoning (BabyVision: 91%)”x.com
- SUPPORT@maxforai“看起来很好，但是要考虑缓存成本”x.com
- SUPPORT@kimmonismus“MiniMax M3.1 Flash matches DeepSeek V4.1 Flash’s off-peak input/output pricing and halves its peak rates”x.com
- SUPPORT@teortaxestex“DeepSeek’s strongest model today has fewer total and much fewer active params than V3 in Dec 2024. Likewise for MiniMax”x.com
- SUPPORT@valsai“M3.1-Flash is solving substantially more problems, but it's also spending far more tokens attempting them”x.com
- SUPPORT@zhihufrontier“M3.1-Flash can produce more polished frontends while taking 2–3× as many agent steps as GLM-5.3 Flash”x.com