CLBench: leaderboard
Knowledge-learning benchmark by Tencent evaluating LLM ability to learn from context across domain knowledge, rule systems, procedural tasks, and pattern discovery. 23 frontier models tested.
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) | 27.9 | #79 (GPT-5.4) |
| 2 | GPT-5.1 (High) | 23.7 | #135 (GPT-5.1) |
| 3 | Hy3-preview | 22.8 | #157 |
| 4 | Grok 4.20 (Reasoning) | 22.2 | #175 (Grok 4.20) |
| 5 | GPT-5.1 | 21.1 | #135 |
| 6 | Claude Opus 4.5 (Thinking) | 21.1 | #83 (Claude Opus 4.5) |
| 7 | Gemini 3.1 Pro (Preview) (High) | 20.8 | #58 (Gemini 3.1 Pro (Preview)) |
| 8 | Claude Opus 4.6 | 20.7 | #63 |
| 9 | Seed 2.0 Pro (High) | 20.6 | #118 (Seed 2.0 Pro) |
| 10 | Qwen 3.6 Plus | 20.3 | #120 |
| 11 | Qwen 3.5 Plus (Thinking) | 19.8 | #125 (Qwen 3.5 Plus) |
| 12 | Kimi K2.5 | 19.3 | #144 |
| 13 | Claude Opus 4.5 | 19.1 | #83 |
| 14 | GLM-5 | 18.7 | #141 |
| 15 | Seed 2.0 Pro (Medium) | 18.7 | #118 (Seed 2.0 Pro) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=clbench · How It Works · Data refreshed daily, snapshot 2026-10-20.