CortexBench - Comparison (Closed-Book): leaderboard
Chinese comparison questions that contrast entities or attributes described in two web documents from associated domains, answered from parametric knowledge alone. Accuracy (%) from a GPT-5.4 semantic-equivalence judge, higher is better; GPT-5.4 also wrote the questions (CORTEX paper, Gan et al., Zhejiang University and JIUTIAN Research, 2026).
Metric: Accuracy (%, GPT-5.4 judge). Source: arxiv.org. Saturation forecast: Around 2028. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 32.3 | #121 |
| 2 | Claude Opus 4.6 | 31.7 | #100 |
| 3 | Gemini 3 Pro | 28.6 | #135 |
| 4 | Claude Sonnet 4.6 | 25 | #158 |
| 5 | DeepSeek V4 Flash | 24.8 | #254 |
| 6 | Gemini 3.1 Pro (Preview) | 24.3 | #91 |
| 7 | DeepSeek V3.2 | 22.6 | #403 |
| 8 | GPT-4o | 21.6 | #741 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=cortexbench-comparison-closed-book · How It Works · Data refreshed daily, snapshot 2026-11-12.