BenchCompass - Context-Grounded Reasoning (Closed Book): leaderboard
Metric: Accuracy (%). Source: arxiv.org. 14 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.5 | 36.5 |
| 2 | Claude Opus 4.7 | 25 |
| 3 | DeepSeek V4 Pro | 16.7 |
| 4 | GLM-5.1 | 15.6 |
| 5 | Gemini 3.1 Pro (Preview) | 14.6 |
| 6 | Qwen 3.5 397B A17B | 14.6 |
| 7 | Gemma 4 31B (IT) | 9.4 |
| 8 | Qwen 3.5 122B A10B | 9.4 |
| 9 | MiniMax-M2.5 | 9.4 |
| 10 | Llama 4 Maverick | 8.3 |
| 11 | Mistral Large 3 | 8.3 |
| 12 | Gemma 4 26B A4B (IT) | 7.3 |
| 13 | Qwen 3 32B | 5.2 |
| 14 | Qwen 3 8B | 1 |
Interactive version: theaggregate.ai/benchmark?slug=benchcompass-context-grounded-reasoning-closed-book · How It Works · Data refreshed daily, snapshot 2026-09-20.