OPEN LINK ↗
// 50d agoBENCHMARK RESULT
MiMo-V2.5-Pro tops open-weights value race
Reddit post argues Xiaomi’s MiMo-V2.5-Pro is now the strongest open-weights model on the author’s Blood on the Clocktower benchmark. The result looks especially compelling on value: strong “good” play, much lower cost, and faster completion than Kimi K2.6, though the benchmark is narrow and other frontier models were not yet tested.
// ANALYSIS
This feels less like a raw SOTA claim and more like a practical win on price-performance: MiMo-V2.5-Pro may be the model you can actually afford to run, not just the one that wins a chart.
- –The asymmetry matters: 88% good-side wins versus 48% evil-side wins suggests strong constructive reasoning but weaker deception/adversarial play
- –The cost delta is the headline for operators: about $0.99/game here versus $2.65/game for Kimi K2.6, with matches finishing in 2-3 hours instead of 10-15
- –A 0.4% tool-call error rate makes the case stronger for agentic workloads, where reliability often matters more than peak benchmark scores
- –This is a highly specialized social-deduction eval, so “best open-weights model” is conditional, not universal
- –The author explicitly hasn’t benched GPT-5.5 Xhigh or Claude Opus 4.7 Max yet, so the top-end closed-model comparison remains open
// TAGS
mimo-v2.5-prollmopen-weightsopen-sourcebenchmarkmoepricingreasoning
DISCOVERED
50d ago
2026-05-01
PUBLISHED
50d ago
2026-05-01
RELEVANCE
9/ 10
AUTHOR
cjami