MiniMax releases M3.1 Flash and says it cuts the cost of agent tasks by 70 percent
The 428-billion-parameter update to M3 runs nearly three times faster, and MiniMax says it was the anonymous model known as Space Bunny.
Chinese AI developer MiniMax has released M3.1 Flash, a faster and cheaper update to the M3 large language model it launched in June. The company says agentic tasks cost 70 percent less to run than on its previous model and that inference is close to three times as fast, with output above 200 tokens per second at everyday peak. The model is about the same size as M3. A developer guide lists 428 billion total parameters, and MiniMax credits the speedup to more aggressive sparsity at inference time. Reasoning is always on, and developers choose one of five effort levels from Low to Max. The model accepts text, image and video input, with prompts of up to 1 million tokens. MiniMax says an early version processed more than 200 trillion tokens while running anonymously as "Space Bunny." In its own testing, the company reports scores of 61.7 on OSWorld V2, a computer-use test, which it says no Chinese model has beaten. It also reports 76.2 on BankerToolBench, 73.7 percent on DeepSWE v1.1 and 82 percent on CyberGym, a security benchmark built by UC Berkeley researchers. These results have not been independently verified. The release is aimed at developers whose agents make dozens of model calls per job and pay for each one.