MiniMax's new AI model scores 13 points higher at under half the cost per task
MiniMax's M3.1 Flash scores 13 points above its last model and costs under half as much per task, but it tries to answer every question, even when unsure.
The Decoder reports that China's MiniMax has released a new AI model, M3.1 Flash. Artificial Analysis, which tests AI models, gives it 42 on its Intelligence Index. That is 13 points above MiniMax's last model, M3.
Artificial Analysis puts the cost at about $0.22 per task, less than half of M3's. Deepseek's V4.1 Flash 1017 costs about $0.26 per task and scores higher overall, 45 to 42. MiniMax's model beats it on some tests, like Humanity's Last Exam (48 percent to 43 percent).
One weak spot: Artificial Analysis gives the model a 93 percent score for making things up. It answers every test question instead of declining when it is unsure. MiniMax plans to publish the model's files for anyone to download on October 23.
From the sourceRead the original at The Decoder →MiniMax now undercuts Deepseek on cost, about $0.22 per task against $0.26, while Deepseek keeps the higher overall score, 45 to 42. Both companies are releasing their model files for anyone to download: Deepseek's are already online and MiniMax's follow on October 23.
The facts
- M3.1 Flash scores 42 on the Artificial Analysis Intelligence Index, 13 points higher than its predecessor M3, at less than half the cost per task.
- Cost per task is about $0.22 for M3.1 Flash versus about $0.26 for Deepseek V4.1 Flash 1017.
- Deepseek V4.1 Flash 1017 leads on the Intelligence Index with 45 points versus MiniMax's 42; Gemini 3.8 Flash scores 41.
- On Humanity's Last Exam, M3.1 Flash scores 48 percent, versus 43 percent for Deepseek.
- On the AA-Omniscience factual accuracy test, M3.1 Flash answers 56 percent of questions correctly, versus 43 percent for Deepseek and 28 percent for GLM-5.3-Flash.
- Artificial Analysis reports a 93 percent hallucination rate for the model, because it tries to answer every test question instead of declining when uncertain.
- MiniMax says the model runs three times faster than M3 and plans an open-weights release on October 23.
- Deepseek's V4.1 Flash 1017 is already available on Hugging Face as an open-weights release.
Sources
- The Decoder: MiniMax M3.1 Flash shows China's AI price race has no floor in sight
Published
More in New models
AmazonNew modelsOfficial releasevia AWS ML blogTwelveLabs' video AI model can now find key moments in long videos on AWS
Amazon Web Services says TwelveLabs' Pegasus 1.5 is now on Amazon Bedrock. It lists matching moments in videos up to two hours long.
DeepSeek says new downloadable AI beats pricier rivals at coding for about 20 cents a job
DeepSeek's own tests show a smaller model beating bigger rivals at coding, but outside tests are missing and China may limit such releases.
AmazonNew modelsAyobi’s pickReflection AI's Beam, a free-to-download coding AI, is now available on Amazon's cloud
Reflection AI released Beam, an AI built to write and fix software over long jobs. Anyone can download and change it, and Amazon now offers it to run on its cloud.