What happened
MiniMax released M3.1 Flash, which scores 42 points on the Artificial Analysis Intelligence Index, 13 points above M3, at about $0.22 per task — less than half the cost.
Why it matters
A 13-point gain at roughly half the cost per task suggests cheaper AI output is arriving quickly, which may pressure rivals' pricing.
What to watch
MiniMax plans to release the model as open weights on October 23.
WHO IT HITSAI developers and product teams choosing model APIs will see a cheaper option at this score level, while rival providers may face pricing pressure to match it.
Summaries like this, in your inbox every morning.
MiniMax's M3.1 Flash arrives as Chinese labs keep pushing down the cost of running a capable reasoning model. The headline result is a 13-point jump on the Artificial Analysis Intelligence Index over M3, paired with a per-task cost of about $0.22 — less than half what M3 needed. That combination, rather than the score alone, is what makes the release notable.
The body positions M3.1 Flash against a set of peers without a clean sweep either way. It ties GLM-5.3-Flash, edges past Gemini 3.8 Flash (41) and trails Deepseek V4.1 Flash 1017 (45) on the index, while on specific tests — 'Humanity's Last Exam' at 48 percent and AA-Omniscience factual accuracy at 56 percent — it beats both Deepseek and GLM. Its 93 percent hallucination rate, which Artificial Analysis attributes to answering every question rather than declining, is the trade-off attached to those gains.
A separate thread is openness. Deepseek's comparable model is already on Hugging Face, and MiniMax says it will follow with open weights on October 23. The API is live now, and MiniMax also offers a desktop app called MiniMax Code.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
Tenstorrent, a 10-year-old Nvidia challenger focused on inference, raised $1.5 billion at a $6.2 billion valua…
PNY is offering NVIDIA RTX PRO 6000 Blackwell Workstation Edition and Max-Q Edition GPUs, configurable with up…
Anthropic's study of its own workforce found 80% of elapsed Claude session time was agents waiting on human in…
The piece lays out IBM's five-type classification by mechanism — simple reflex, model-based reflex, goal-based…
AINOW published a comparison of 17 AI agents as of October 2026, covering pricing, free-trial availability, an…
The Nikkei column says the scramble for capital it flagged in June has become fiercer, with the 10-year US Tre…
