BrainBench (Commonsense) - Consistency: leaderboard
Metric: Consistency (%): the share of the 100 English BrainBench questions answered correctly in all 10 runs (the paper's reliability metric), the 100 BrainBench questions (20 categories of commonsense reasoning traps, 5 questions each), every question sent 10 times as a standalone zero-shot prompt with no system prompt, a judge LLM marking each response's conclusion against the ground truth; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 (Thinking) | 74 | #60 (Claude Opus 4.6) |
| 2 | Claude Opus 4.6 | 71 | #60 |
| 3 | Claude Sonnet 4.6 | 69 | #85 |
| 4 | GPT-5.4 (Thinking) | 64 | #76 (GPT-5.4) |
| 5 | GPT-5.4 | 63 | #76 |
| 6 | Claude Haiku 4.5 | 58 | #271 |
| 7 | GPT-4o | 27 | #333 |
| 8 | GPT-4o Mini | 24 | #588 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=brainbench-commonsense-consistency · How It Works · Data refreshed daily, snapshot 2026-10-11.