ChLogic - English Difficult: leaderboard
English binary logical-judgment accuracy on ChLogic’s Difficult aligned set (2,000 items), reported in Table 2. Each model’s Difficult score is retained separately from the other set and from Chinese or back-translated formulations. Seven native model labels are retained; unspecified checkpoint and thinking settings are not inferred.
Metric: English logical-judgment accuracy (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 32B | 96.05 | #449 |
| 2 | Qwen 3 14B | 93.02 | #542 |
| 3 | GLM-5.1 | 84.7 | #109 |
| 4 | Qwen 3 8B | 80.5 | #678 |
| 5 | Ministral 3B | 73.3 | #1273 |
| 6 | Ministral 8B | 68.55 | #1036 |
| 7 | Qwen 3 0.6B | 55 | #1448 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=chlogic-english-difficult · How It Works · Data refreshed daily, snapshot 2026-11-05.