Clembench Text v3.0: leaderboard
Clembench Text v3.0 evaluates chat-optimized language models as conversational agents in language games. The input modality is text. Its pinned metric is clemscore, a leaderboard score reported in the source results table alongside related task-level measures. Source: BenchmarkList’s attributed result table.
Metric: clemscore. Source: benchmarklist.com. 31 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 (High) | 90.1 | #148 (Claude Sonnet 4.5) |
| 2 | Claude Sonnet 4.5 | 87.42 | #148 |
| 3 | GPT-5.2 (High) | 84.19 | #115 (GPT-5.2) |
| 4 | Gemini 3 Flash | 84.03 | #103 |
| 5 | GPT-5.2 | 81.66 | #115 |
| 6 | GPT-5.2 (Medium) | 79.61 | #115 (GPT-5.2) |
| 7 | Kimi K2 (Thinking) | 77.79 | #252 (Kimi K2) |
| 8 | Kimi K2.5 (Non-reasoning) | 60.28 | #152 (Kimi K2.5) |
| 9 | Llama 3.3 70B Instruct | 50 | #541 |
| 10 | Qwen 2.5 72B Instruct | 48.07 | #469 |
| 11 | Llama 3.1 70B Instruct | 46.8 | #567 |
| 12 | Qwen 3 Next 80B A3B Instruct | 45.24 | #381 |
| 13 | Qwen 2.5 Coder 32B Instruct | 35.32 | #534 |
| 14 | Llama 3.1 8B Instruct | 25.28 | #1017 |
| 15 | aya-expanse-32B | 16.9 | #687 |
Interactive version: theaggregate.ai/benchmark?slug=clembench-text-v3-0 · How It Works · Data refreshed daily, snapshot 2026-11-05.