CLBench - Empirical Discovery & Simulation: leaderboard
CL-bench (context learning) category 'Empirical Discovery & Simulation': tasks that require discovering patterns from empirical data or simulations supplied in a professional context the model has to learn from. Solving rate in percent (published as a 0-1 fraction), higher is better; labels keep the reasoning effort the source records.
Metric: Solving Rate (%). Source: www.clbench.com. Saturation forecast: Around 2030. 36 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 (xHigh) | 22.8 | #83 (GPT-5.4) |
| 2 | Hy3-preview | 19.6 | #166 |
| 3 | Kimi K2.5 | 18.1 | #152 |
| 4 | GPT-5.1 (High) | 18.1 | #145 (GPT-5.1) |
| 5 | Seed 2.0 Pro (High) | 17.2 | #122 (Seed 2.0 Pro) |
| 6 | Qwen 3.5 Plus (Thinking) | 16.7 | #133 (Qwen 3.5 Plus) |
| 7 | Claude Opus 4.5 (Thinking) | 15.1 | #87 (Claude Opus 4.5) |
| 8 | Gemini 3.1 Pro (Preview) (High) | 14.5 | #64 (Gemini 3.1 Pro (Preview)) |
| 9 | Seed 2.0 Pro (Medium) | 13.9 | #122 (Seed 2.0 Pro) |
| 10 | O3 (High) | 13.7 | #132 (O3) |
| 11 | GPT-5.1 | 13.6 | #145 |
| 12 | Qwen 3.6 Plus | 13.5 | #124 |
| 13 | Kimi K2 (Thinking) | 12.6 | #252 (Kimi K2) |
| 14 | Grok 4.20 (Reasoning) | 12.3 | #179 (Grok 4.20) |
| 15 | GPT-5.2 | 12.1 | #115 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=clbench-empirical-discovery-simulation · How It Works · Data refreshed daily, snapshot 2026-11-05.