MiniMax Releases M3.1 Flash Model with Faster Inference and Lower Costs

On October 19, 2026, MiniMax officially released its new-generation Flash model, MiniMax M3.1 Flash. The company revealed that Space Bunny, an anonymous model that processed over 200 trillion tokens, was an early version of this model.

The new model delivers generational improvements in agentic tasks, coding, computer use, and cybersecurity, reaching leading-model performance in complex agent tasks, software development, and security benchmarks. Online inference speed is nearly 3x faster, while usage costs are reduced by 70%.

Pricing is set at 1 yuan per million input tokens, 4 yuan per million output tokens, and 0.1 yuan per million cached tokens. According to official comparisons, M3.1 Flash performs on par with DeepSeek V4.1 Flash-1017, with a better Pareto curve in per-task cost-effectiveness, though it trails slightly in the coding-focused DeepSWE benchmark.

M3.1 Flash is now available via the MiniMax API, MPlan, and MiniMax Code, with weights to be open-sourced later.