MiniMax has officially launched MiniMax M3.1 Flash, a next-generation model it describes as delivering "a generational leap" in agentic tasks, coding, Computer Use and cybersecurity. The company announced the rollout on its blog, alongside claims that online inference is now nearly three times faster while usage costs drop by 70%.
MiniMax says an early, anonymous version of the model — nicknamed "Space Bunny" — processed more than 200 trillion tokens before today's release. The model is available now through the MiniMax API, the M Plan subscription and MiniMax Code, with model weights promised for a later date, expected around October 23.
MiniMax M3.1 Flash specs and pricing
According to MiniMax, M3.1 Flash reaches the level of leading models on benchmarks covering complex agentic tasks, computer use, software development and cybersecurity. The company cites Automation-Bench for agentic task execution, DeepSWE for real-world coding and Agents' Last Exam (ALE) for economically valuable computer-use tasks.
Efficiency is a central part of the pitch. MiniMax says it has further optimized inference through "more aggressive sparsity," allowing output speeds of 200+ tokens per second (TPS) in everyday use, with median P50 speeds consistently reaching 170+ TPS. For agentic tasks, the company claims overall usage cost is 70% lower than the previous model.
- Claimed generational leap in agentic tasks, coding, Computer Use and cybersecurity
- Online inference nearly 3x faster, usage costs down 70% versus predecessor
- Output speeds up to 200+ TPS, with median P50 speeds of 170+ TPS
