MiniMax M3.1 Flash benchmark table leaks: tops automation evals at 62.8, size undisclosed
QuixiAI · x · 2026-10-20
QuixiAI shared a multi-benchmark comparison table of MiniMax M3.1 Flash against DeepSeek-V4.1-Flash, Kimi-K3, GLM-5.3, GPT-5.6 and Opus 5. The model leads most CoWork automation benchmarks (62.8 Automation-Bench, 76.2 BankerToolBench, 60.3 LOCA-Bench 256K), scores 82.2 on CyberGym and 82.5 on CharXiv RQ, while trailing DeepSeek on DeepSWE coding (73.7 vs 77.0). The table includes detailed eval harness settings, and the author notes MiniMax hasn't disclosed the model's size.
Related event: MiniMax Releases M3.1 Flash, Topping DeepSeek in Automated Benchmarks(2 posts)→
More from Models
- User swaps Copilot 365 sessions to Claude Opus 5.5 and stuck sessions vanish — koltregaskes · 2026-10-20
- Stanford Zou Lab launches Collaborative Intelligence Leaderboard; OpenAI models rank best teammates — james_y_zou · 2026-10-20
- Researcher criticizes missing ablations in new model tech report, citing Llama precedent — eliebakouch · 2026-10-20
- Benchmaxxing score flags models ranking higher on public benchmarks than held-out ones — airesearch12 · 2026-10-20
- DeepSeek model catches its own backup mistake, says "oops," and invents a lateral fix — EffectiveMedium2683 · 2026-10-20
- Exa 3.0 ties Perplexity at #1 on Artificial Analysis Search Index with score of 80 — yoimnotkesku · 2026-10-20