LongBench Pro - Hard - Non-Thinking: leaderboard
Evaluates long-context understanding across English and Chinese tasks at four difficulty levels. Non-thinking and thinking results use separate published columns; missing settings remain absent. This series reports Hard - Non-Thinking using Score (%). Source: the original paper result table.
Metric: Score (%). Source: arxiv.org. Saturation forecast: Around February 2027. 34 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Flash | 57.87 | #255 |
| 2 | Claude Sonnet 4 | 57.57 | #207 |
| 3 | Qwen 3 235B A22B 2507 Instruct | 52.34 | #315 |
| 4 | DeepSeek V3.2 | 51.63 | #210 |
| 5 | Qwen 3 Next 80B A3B Instruct | 49.93 | #381 |
| 6 | DeepSeek V3.1 | 49.29 | #275 |
| 7 | DeepSeek V3 (0324) | 48.3 | #349 |
| 8 | Claude 3.7 Sonnet | 47.29 | #259 |
| 9 | GPT-4o | 44.88 | #357 |
| 10 | Kimi K2 0905 | 43.43 | #292 |
| 11 | GLM-4.6 | 43.07 | #265 |
| 12 | Qwen 3 30B A3B 2507 Instruct | 41.32 | #492 |
| 13 | GLM-4.5 | 40.21 | #283 |
| 14 | MiniMax-Text-01 | 38.67 | #519 |
| 15 | Qwen 3 32B | 38.61 | #449 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=longbench-pro-hard-non-thinking · How It Works · Data refreshed daily, snapshot 2026-11-05.