# Clocktower Bench leaderboard

external benchmark

> Every figure below is reproduced as its upstream source published it: nothing is modelled, estimated, interpolated or converted. `Unavailable` means no source published the value — it is never a zero. Each value carries its evidence state and the date it was observed.

Page: https://modelscale.dev/leaderboards/benchmarks/external-clocktower-bench

Clocktower Bench (Blood on the Clocktower). Every model the catalog carries a published Clocktower Bench value for, ranked by that value.

## Benchmark record

| Field | Value |
| --- | --- |
| Category | external |
| Measure | Relative rating from side-swapped game pairs |
| Tasks | 386 rated head-to-head games of Blood on the Clocktower (10 to 12 players) |
| Difficulty | Social deduction, deception and multi-agent reasoning |
| Published by | [Clocktower Bench leaderboard](https://clocktowerbench.com/) |

Frontier models play the social deduction game Blood on the Clocktower against each other, one model as the town and one as the Demon's team. The rating combines catching liars (detection) with lying well (deception).

## Clocktower Bench ranking

12 models with a published Clocktower Bench value, ordered by that value, highest first. Models the source has not scored on this benchmark are not listed — they are unmeasured, not last.

| Rank | Model | Provider | Relative rating from side-swapped game pairs | Evidence |
| --- | --- | --- | --- | --- |
| 1 | [GPT-6 Astra](https://modelscale.dev/models/gpt-6-astra) | OpenAI | 1859 | Observed 2026-11-21 · source benchlm:models |
| 2 | [GPT-6.1 Sol](https://modelscale.dev/models/gpt-6-1-sol) | OpenAI | 1734 | Observed 2026-11-21 · source benchlm:models |
| 3 | [Claude Fable 5.1](https://modelscale.dev/models/claude-fable-5-1) | Anthropic | 1595 | Observed 2026-11-21 · source benchlm:models |
| 4 | [GPT-5.6 Sol](https://modelscale.dev/models/gpt-5-6-sol) | OpenAI | 1591 | Observed 2026-11-21 · source benchlm:models |
| 5 | [Claude Opus 5.5](https://modelscale.dev/models/claude-opus-5-5) | Anthropic | 1549 | Observed 2026-11-21 · source benchlm:models |
| 6 | [Claude Fable 5](https://modelscale.dev/models/claude-fable-5) | Anthropic | 1544 | Observed 2026-11-21 · source benchlm:models |
| 7 | [GPT-5.5](https://modelscale.dev/models/gpt-5-5) | OpenAI | 1482 | Observed 2026-11-21 · source benchlm:models |
| 8 | [Claude Opus 4.6](https://modelscale.dev/models/claude-opus-4-6) | Anthropic | 1432 | Observed 2026-11-21 · source benchlm:models |
| 9 | [GPT-5.4](https://modelscale.dev/models/gpt-5-4) | OpenAI | 1276 | Observed 2026-11-21 · source benchlm:models |
| 10 | [Claude Opus 4.7](https://modelscale.dev/models/claude-opus-4-7) | Anthropic | 1241 | Observed 2026-11-21 · source benchlm:models |
| 10 | [Claude Opus 4.7 (Adaptive)](https://modelscale.dev/models/claude-opus-4-7-max) | Anthropic | 1241 | Last good 2026-11-18 · source benchlm:models |
| 12 | [Claude Opus 4.8](https://modelscale.dev/models/claude-opus-4-8) | Anthropic | 1198 | Observed 2026-11-21 · source benchlm:models |

Rows are ordered by the value the source published, highest first. The source does not state whether a higher value is the better result, so this document does not either: for a benchmark that measures a rate of failure, read the table from the bottom.
