AA Long Context Reasoning: leaderboard
Each of the 100 questions comes with a set of long documents totalling about 100,000 tokens, such as company reports, academic papers, government consultations, legal texts and industry reports, and the answer has to be reasoned out from passages spread across them. Another model judges whether the response matches the official answer, and the board reports the pass rate averaged over three runs. Higher is better.
Metric: Accuracy (%). Source: artificialanalysis.ai. Source: benchmarklist.com. Saturation forecast: Around January 2027. 579 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K3 (Max) | 88.67 | #30 (Kimi K3) |
| 2 | Step 5 Preview | 88.33 | #44 |
| 3 | MiMo-V2.6-Pro | 86.33 | #42 |
| 4 | GPT-5.5 (xHigh) | 84.33 | #31 (GPT-5.5) |
| 5 | GPT-5.5 (High) | 84.33 | #31 (GPT-5.5) |
| 6 | GPT-5.6 Sol (Max) | 84 | #19 (GPT-5.6 Sol) |
| 7 | DeepSeek V4.1 Flash (Max) | 84 | #36 (DeepSeek V4.1 Flash) |
| 8 | GPT-6.1 Sol (Low) | 84 | #7 (GPT-6.1 Sol) |
| 9 | Gemini 3.8 Flash (Medium) | 84 | #21 (Gemini 3.8 Flash) |
| 10 | GPT-5.6 Luna (Max) | 83.67 | #68 (GPT-5.6 Luna) |
| 11 | GPT-6 Sol (Max) | 83.67 | #14 (GPT-6 Sol) |
| 12 | GPT-6 Sol (High) | 83.67 | #14 (GPT-6 Sol) |
| 13 | Muse Wonder Full (Max) | 83.67 | #10 (Muse Wonder Full) |
| 14 | GPT-6 Luna (Max) | 83.33 | #56 (GPT-6 Luna) |
| 15 | Muse Glimmer 30B (High) | 83.33 | #178 (Muse Glimmer 30B) |
No result here: #3 Claude Opus 5.5, #9 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=aa-long-context-reasoning · How It Works · Data refreshed daily, snapshot 2026-10-20.