MiniMax opens M3.1 Flash, the Space Bunny model
Why it matters
MiniMax made the weights of M3.1 Flash public on Hugging Face on Oct 23, and says the mystery "Space Bunny" model, which topped global leaderboards with more than 200 trillion tokens, was an earlier version of it. It is a mixture-of-experts model with about 428B total parameters and 23B active per token; it takes text, image and video in, returns text, handles up to a 1M-token context, and is built for coding, agentic workflows, computer use and cybersecurity. In MiniMax's own comparison table against GPT-5.6 Sol and Claude Opus 5 (a few cells come from third-party leaderboards), it scores 76.2 on BankerToolBench, tied with Opus 5, and 73.7% on DeepSWE v1.1, while Opus 5 still leads Terminal Bench 4.0 (51.8 vs 38.1) and HLE without tools (55.6 vs 43.4). API pricing is $0.15 / $0.60 per 1M input/output tokens; MiniMax says that is 1/3 the cost of M3 at 3x the speed, helped by sparse attention throughout, quantization-aware training and DSpark speculative decoding (draft head included). Caveats: these are MiniMax's published numbers, and the MiniMax Community License is free for non-commercial use only; commercial use must display "Built with MiniMax M3.1-Flash" and notify MiniMax, and products with more than $20M in yearly revenue need prior written authorization.
Read the source: MiniMax (opens in a new tab)
All hits from 23 Oct 2026More Open ModelsGet the daily hits by email