Domestic model Agent ranks first! MiniMax dramatically upgrades Flash's combat power, all-day big fat fish valley price

What happened: MiniMax officially released M3.1 Flash, the model previously tested anonymously on OpenRouter as Space Bunny. It tops domestic models on the Agents' Last Exam (ALE) benchmark and ranks in the global first tier on coding, cybersecurity, and Computer Use tests. Key numbers: ALE 29.6%, AA 43 points, DeepSWE v1.1 73.7%, CyberGym 82.2, BankerToolBench Why it matters: The content frames M3.1 Flash and DeepSeek V4.1 Flash as complementary low-cost domestic models, with MiniMax stronger on cross-SaaS agent workflows and financial tool calls, pressuring closed-source flagship pricing.

MiniMax officially released M3.1 Flash, the model previously tested anonymously on OpenRouter as Space Bunny. It tops domestic models on the Agents' Last Exam (ALE) benchmark and ranks in the global first tier on coding, cybersecurity, and Computer Use tests.

The content frames M3.1 Flash and DeepSeek V4.1 Flash as complementary low-cost domestic models, with MiniMax stronger on cross-SaaS agent workflows and financial tool calls, pressuring closed-source flagship pricing.