DiscoBench - Neutral - Accuracy: leaderboard
Metric: End-to-end accuracy (%). Source: arxiv.org. 11 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) (High) | 40.8 |
| 2 | Claude Opus 4.7 (Max) | 39.8 |
| 3 | DeepSeek V4 Pro (Max) | 35.5 |
| 4 | Kimi K2.6 (Thinking) | 29.4 |
| 5 | GLM-5.1 (Thinking) | 28.4 |
| 6 | GPT-5.4 (xHigh) | 27.5 |
| 7 | MiMo V2.5 Pro (Thinking) | 24.2 |
| 8 | Hy3-preview (High) | 16.1 |
| 9 | Qwen 3.6 Max (Preview) (Thinking) | 12.3 |
Interactive version: theaggregate.ai/benchmark?slug=discobench-neutral-accuracy · How It Works · Data refreshed daily, snapshot 2026-09-19.