Logical Consistency Reformulation - Base Accuracy: leaderboard
Zero-shot accuracy on the 350 original CRTBench questions. GPT reasoning-effort none and high runs remain distinct; this measures base accuracy, not consistency across reformulations.
Metric: Accuracy (%, zero-shot base questions). Source: arxiv.org. Saturation forecast: Estimated already saturated. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 (Non-reasoning) | 98.9 | #79 (GPT-5.4) |
| 2 | GPT-5.4 Mini (Non-reasoning) | 98.9 | #212 (GPT-5.4 Mini) |
| 3 | O4 Mini | 98.6 | #180 |
| 4 | Gemini 2.5 Flash | 96.9 | #245 |
| 5 | GPT-5.4 Mini (High) | 96 | #212 (GPT-5.4 Mini) |
| 6 | Claude Sonnet 4 | 95.1 | #199 |
| 7 | GPT-5.4 (High) | 92.9 | #79 (GPT-5.4) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=logical-consistency-reformulation-base-accuracy · How It Works · Data refreshed daily, snapshot 2026-10-20.