MiniMax launches M3.1 flash with 3x speed boost and 70% cost cut

On October 19, 2026, MiniMax announced the release of its M3.1 Flash model, which offers a 3x increase in inference speed and a 70% reduction in usage costs. The model, previously tested under the anonymous Space Bunny project with over 200T Token processing, now delivers enhanced performance in agentic tasks, programming, computer use, and cybersecurity. It achieved top scores in benchmarks like AutomationBench-AA, BankerToolBench, and DeepSWE v1.1, demonstrating strong capabilities in complex workflows and software development. M3.1 Flash also excelled in OSWorld V2 and Agents' Last Exam (ALE), ranking first among domestic models. The model’s cost structure includes $1 per million input tokens, $4 per million output tokens, and $0.1 per million for cached reads. MiniMax plans to release the model’s weights and expand access through its API and MPlan platform in the near future.