Vals Index: M3.1 Flash ranks #5 among open-weight models, biggest gains in coding
Independent evaluator ValsAI reports M3.1 Flash is #5 among open-weight models on the Vals Index, one point behind MiMo V2.6 Flash, improving over M3 on every benchmark with the largest gains in coding and smaller ones in tax and finance.
MiniMax M3.1-Flash is now #5 among open-weight models on the Vals Index, just one point behind MiMo V2.6 Flash. M3.1-Flash improved over Minimax-M3 on all benchmarks, but the biggest gains were on coding tasks, with smaller jumps in tax and finance.
M3.1-Flash also made meaningful progress on Vibe Code Bench, improving by 26.4 points and climbing from #64 to #38 overall. The number of apps receiving a zero score fell from 15 to just 3, while the number of perfect apps doubled from 6 to 12. The improvement wasn't limited to polishing already successful apps. M3.1-Flash was substantially less likely to fail outright.
One notable change is how much longer M3.1-Flash reasons through problems. On Tax Agent and Legal Research, it generated roughly 15 times as many output tokens as M3, most of them devoted to reasoning. That additional reasoning likely explains part of the improvement. M3.1-Flash is solving substantially more problems, but it's also spending far more tokens attempting them. The gains in accuracy shouldn't be confused with gains in efficiency.