MiniMax M3.1 Flash released, tops Agent task evaluation
MiniMax released M3.1 Flash and claimed the previously anonymous model Space Bunny; it ranked first in the Automation-Bench complex Agent task comparison, and tied for first with Opus 5 on BankerToolBench at 76.2%. Leiphone used the same 3D campus building task for a real-world test: MiniMax Code plus M3.1 Flash took 55.4 minutes, while OpenCode plus DeepSeek V4.1 Flash took about 121 minutes, with model costs for both around 6.5 yuan. Pricing is 1 yuan per million input Tokens, 4 yuan for output, and 0.1 yuan for cache reads, roughly half the input and output prices of the previous generation M3, with no peak/off-peak distinction all day.
Why read this
SelectedAI score 83
The article compares the leaderboard results of two domestic Flash models with the same Agent real-world test, revealing the trade-offs of low-cost models on complex tasks.
Topics
Body
The English body is being prepared. You can open the original source.