PatternEval (Thinking): leaderboard
PatternEval (2026) collects 2,415 multimodal prompts on which model responses often fail in user-visible ways, and runs hybrid-thinking models once with thinking on and once with it off. This board is thinking inference; the score is the accuracy of the final answer under each source task's criteria as judged by Qwen3-Max, in percent; higher is better.
Metric: Accuracy (%; correctness of the final user-visible answer under the source task's criteria, judged by Qwen3-Max; thinking inference; PatternEval: 2,415 failure-enriched image-prompt pairs in three task families (visual perception and grounding, OCR and structured images, multimodal knowledge reasoning); official default decoding with only the reasoning control changed). Source: arxiv.org. Saturation forecast: Around February 2027. 25 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 74.9 | #49 |
| 2 | Kimi K2.6 (Thinking) | 73.19 | #178 (Kimi K2.6) |
| 3 | Qwen 3.7 Max (Thinking) | 72.3 | #145 (Qwen 3.7 Max) |
| 4 | GPT-5.4 (Thinking) | 69.54 | #123 (GPT-5.4) |
| 5 | Qwen 3.5 397B A17B (Thinking) | 68.93 | #246 (Qwen 3.5 397B A17B) |
| 6 | Claude Opus 4.8 (Thinking) | 67.7 | #56 (Claude Opus 4.8) |
| 7 | Claude Opus 4.6 (Thinking) | 67.63 | #102 (Claude Opus 4.6) |
| 8 | Kimi K2.5 (Thinking) | 67.06 | #256 (Kimi K2.5) |
| 9 | Qwen 3.5 122B A10B (Thinking) | 66.3 | #331 (Qwen 3.5 122B A10B) |
| 10 | Qwen 3.5 35B A3B (Thinking) | 65.07 | #482 (Qwen 3.5 35B A3B) |
| 11 | Qwen 3.5 27B (Thinking) | 64.05 | #359 (Qwen 3.5 27B) |
| 12 | Qwen 3.6 35B A3B (Thinking) | 63.9 | #462 (Qwen 3.6 35B A3B) |
| 13 | Qwen 3.6 27B (Thinking) | 63.72 | #341 (Qwen 3.6 27B) |
| 14 | Qwen 3.5 9B (Thinking) | 59.2 | #803 (Qwen 3.5 9B) |
| 15 | Qwen 3 VL 32B (Thinking) | 55.75 | #559 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=patterneval-thinking · How It Works · Data refreshed daily, snapshot 2026-11-13.