LongBench Pro - Extreme - 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 Extreme - Thinking using Score (%). Source: the original paper result table.
Metric: Score (%). Source: arxiv.org. Saturation forecast: Around September 2027. 46 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 50.77 | #159 |
| 2 | GPT-5 | 48.37 | #105 |
| 3 | Gemini 2.5 Flash | 47.39 | #255 |
| 4 | Claude Sonnet 4 | 47.05 | #207 |
| 5 | DeepSeek V3.2 | 44.27 | #210 |
| 6 | Qwen 3 235B A22B 2507 (Thinking) | 43.39 | #263 |
| 7 | Qwen 3 235B A22B 2507 Instruct | 43.24 | #315 |
| 8 | DeepSeek V3.1 | 42.68 | #275 |
| 9 | Qwen 3 Next 80B A3B (Thinking) | 42.47 | #312 |
| 10 | DeepSeek R1 0528 | 41.49 | #232 |
| 11 | DeepSeek R1 | 40.76 | #262 |
| 12 | Qwen 3 Next 80B A3B Instruct | 40.47 | #381 |
| 13 | Qwen 3 30B A3B 2507 (Thinking) | 40.47 | #379 |
| 14 | Claude 3.7 Sonnet | 40.07 | #259 |
| 15 | GLM-4.6 | 38.88 | #265 |
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-extreme-thinking · How It Works · Data refreshed daily, snapshot 2026-11-05.