Chinese AI company MiniMax introduced M3.1 Flash, a reasoning AI model scoring 42 on the Artificial Analysis Intelligence Index. The model achieves higher performance than its predecessor at less than half the cost per task ($0.22), while operating three times faster, according to the company.
On specialized evaluations, M3.1 Flash achieved 48 percent on Humanity’s Last Exam and 56 percent on the AA-Omniscience test for factual accuracy, outperforming rival models in select metrics. However, testing by Artificial Analysis revealed a 93 percent hallucination rate due to the model attempting answers rather than declining when uncertain.
Featuring a 1-million-token context window and multimodal processing capabilities, M3.1 Flash is available via API at $0.15 per million input tokens and $0.60 per million output tokens. MiniMax plans to release open weights for the model on October 23.
Why it matters
Aggressive pricing and open-weights strategies from Chinese AI labs continue driving down inference costs globally.
High hallucination rates in fast reasoning models underline ongoing trade-offs between speed, cost, and factual reliability.
Source: the-decoder.com



