MiniMax ships M3.1 Flash model aimed at agent workloads

MiniMax released M3.1 Flash on October 19. It lists input at $0.15 per million tokens, output at $0.60 per million, and cached input at $0.015 per million. The model supports a one-million-token context window.

MiniMax states inference runs nearly three times faster than M3. It claims a 70 percent reduction in agent costs, though the figure depends on token usage per completed job rather than the rate card alone.

Independent tests from Artificial Analysis show a 57 percent drop in average cost per Intelligence Index task. The model scores 42 on that index and 73.7 percent on DeepSWE v1.1 coding, close to several frontier models.

The weights are scheduled for release on October 23. The license remains unpublished. A stronger Pro variant is also planned but has no announced date.