ReLE - Overall: leaderboard
Metric: Accuracy (%). Source: nonelinear.com. 162 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 3.7 Max | 76.9 |
| 2 | Claude Opus 5 | 76.8 |
| 3 | Gemini 3.8 Flash | 76.2 |
| 4 | Kimi K3 | 75.6 |
| 5 | Qwen 3.8 Max | 75.4 |
| 6 | Gemini 3.7 Flash | 75.4 |
| 7 | Qwen 3.6 Max Preview | 75.4 |
| 8 | GPT-5.5 | 75.3 |
| 9 | Gemini 3.1 Pro (Preview) | 75.2 |
| 10 | GLM-5.3 | 75.2 |
| 11 | DeepSeek V4 Pro | 74.9 |
| 12 | Claude Opus 4.8 (Thinking) | 74.7 |
| 13 | DeepSeek V4.1 Flash | 74.4 |
| 14 | GPT-5.6 Pro Sol | 74.4 |
| 15 | GLM-5.3 Flash | 74.2 |
Interactive version: theaggregate.ai/benchmark?slug=rele-overall · How It Works · Data refreshed daily, snapshot 2026-09-19.