MiniMax has released M3.1 Flash, a new reasoning SI model that demonstrates the relentless downward pressure on costs in China's competitive SI landscape. The model scores 42 points on the Artificial Analysis Intelligence Index, a 13-point jump from its predecessor M3, while reducing the cost per task to approximately $0.22.

What Happened

According to Artificial Analysis, M3.1 Flash ties with GLM-5.3-Flash and surpasses Gemini 3.8 Flash (41 points) on the Intelligence Index, though it trails Deepseek V4.1 Flash 1017 (45 points). The model shows significant gains in complex reasoning, achieving 48 percent on "Humanity's Last Exam," which outperforms both Deepseek and GLM. On the CritPt benchmark, which tests research-level physics reasoning across 71 complex problems, M3.1 Flash scores 22 percent, placing it ahead of GLM-5.3-Flash (15 percent) and Gemini 3.8 Flash (18 percent), but just behind Deepseek (24 percent). In factual accuracy tests on AA-Omniscience, the model answers 56 percent of questions correctly, significantly higher than GLM-5.3-Flash (28 percent) and Deepseek (43 percent).

The cost efficiency is driven by a reduction in output tokens. MiniMax reports the model runs three times faster than M3 and consumes about 189 million output tokens across benchmarks, compared to Deepseek's 224 million. This efficiency lowers the cost per task to $0.22, undercutting Deepseek's $0.26. However, Artificial Analysis notes a high hallucination rate of 93 percent, attributed to the model's tendency to answer every test question rather than declining when uncertain.

M3.1 Flash supports up to one million tokens of context and accepts both text and image inputs. API pricing is set at $0.15 per million input tokens and $0.60 per million output tokens, with cached inputs at $0.015. The API is live at platform.minimax.io, and MiniMax also offers a desktop app called MiniMax Code. MiniMax plans to release the model as open weights on October 23, following Deepseek's recent open-weight release on Hugging Face. On DeepSWE v1.1, MiniMax-M3.1-Flash scores up to 73.7 percent at under $0.30 per task.

Why It Matters

The release of M3.1 Flash underscores the intensifying price war among Chinese SI labs, where performance gains are increasingly coupled with aggressive cost reductions. By offering a model that matches or exceeds some capabilities of frontier competitors at a fraction of the cost, MiniMax is challenging the notion that high-quality reasoning requires premium pricing. This dynamic forces developers and enterprises to reconsider budget allocations for SI tools, particularly for tasks requiring high factual accuracy and complex reasoning. The open-weight release strategy also suggests a broader industry trend toward democratizing access to advanced SI models, potentially accelerating adoption in cost-sensitive markets.

The Bottom Line

MiniMax M3.1 Flash delivers a 13-point improvement on the Intelligence Index and leads in several reasoning benchmarks at a lower cost than key rivals like Deepseek. While it exhibits a high hallucination rate, its efficiency and upcoming open-weight availability position it as a significant option in the competitive SI market. The launch highlights how Chinese SI companies are driving down prices while rapidly advancing model capabilities.