MiniMax releases M3.1 Flash reasoning model via API at $0.15/$0.60 per million tokens, open weights planned for Oct 23

MiniMax has released M3.1 Flash, a reasoning model with a 1M-token context window that accepts text and image input. The API is live at platform.minimax.io, priced at $0.15 per million input tokens, $0.60 per million output tokens and $0.015 for cached input. Artificial Analysis, an independent tester, scores it 42 on its Intelligence Index, up from 29 for M3, at about $0.22 per task. MiniMax says open weights will follow on 23 October, and it also offers a desktop app called MiniMax Code.

Developers and enterprises running high-volume workloads can test a cheap, fairly strong reasoning model today, and could use it to put pressure on pricing from other suppliers. The planned open-weights release would let teams self-host it, but only if the licence and hardware needs suit them. The gain over M3 is large, but it only ties GLM-5.3-Flash and trails Deepseek V4.1 Flash overall, so it is a competitive option rather than a clear leader.

The headline claim that the price race has "no floor" is an extrapolation from one release, not evidence. The article also glosses over the 93% hallucination rate on AA-Omniscience, which matters a lot for any use where wrong answers are costly. The benchmark ranking is a single index from one tester, and rankings among Flash-tier models shift within weeks.

1. 01Whether the open weights actually appear on 23 October, and under what licence (commercial use, restrictions, size and hardware needed to run them).
2. 02Independent reruns of the benchmarks and real-task tests, especially hallucination and long-context accuracy at the full 1M tokens, compared with MiniMax's own claims.
3. 03Whether Deepseek, Zhipu (GLM) or Google cut prices or ship updated Flash models within the next month, and whether Artificial Analysis cost-per-task figures hold up at scale.