Blood on the Clocktower benchmark: Kimi K3.1 Pro tops open models, just below GPT 5.5
eliebakouch · x · 2026-11-12
eliebakouch added 5 open models to Dwarkesh Patel's Blood on the Clocktower social-deduction benchmark, watching their chains of thought as they lie, bluff and catch each other. Kimi K3.1 Pro is the best open model, landing mid-pack among closed models just below GPT 5.5, with a coherent open-model ranking (DeepSeek V4.1 Flash beats its older version, GLM 5.3 beats 5.3 Flash). A notable finding: models good at the game are also good at spotting villains from just the public transcript — useful for cheaper evaluation.
More from Models
- VulnCheck audit: only 0.57% of Anthropic Glasswing's 37,966 findings ever fixed — WeldPond · 2026-11-12
- On PostTrainBench, AI post-training now beats the human-tuned SmolLM3-3B from the same base — maksym_andr · 2026-11-12
- Claude flags mispronunciations in video audio it claims it can't hear — RileyRalmuto · 2026-11-12
- Heavy Users Say Claude's Safety Rails Block Even Basic Chemistry Work, Forcing Fallback to Opus — rbhar90 · 2026-11-12
- Community teams release wave of abliterated Qwen models for 8–24GB local GPUs — udmrzn · 2026-11-12
- Redditor claims Qwen 4 Max slots between DeepSeek v4.1 and Kimi 3.1 Pro in output quality — power97992 · 2026-11-12