PatternEval - Response-Pattern Failure Rate (Non-Thinking): leaderboard
Metric: Bad-pattern trigger rate (%, lower is better; share of final responses showing at least one of chain-of-thought leakage, repetition, logical contradiction or performative reasoning, labelled by a Seed-2.0-Pro judge with image access; non-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 December 2026. 25 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.5 (Non-reasoning) | 5.04 |
| 2 | GPT-5.4 (Non-reasoning) | 5.34 |
| 3 | Seed 2.0 Pro (Non-reasoning) | 6.94 |
| 4 | Claude Opus 4.8 | 22.06 |
| 5 | Qwen 3.7 Max (Non-reasoning) | 26.77 |
| 6 | Qwen 3.7 Plus (Non-reasoning) | 27.82 |
| 7 | Qwen 3.5 122B A10B (Non-reasoning) | 28.45 |
| 8 | Qwen 3.5 27B (Non-reasoning) | 29.32 |
| 9 | Qwen 3.5 397B A17B (Non-reasoning) | 29.32 |
| 10 | Qwen 3.6 35B A3B (Non-reasoning) | 32.8 |
| 11 | Qwen 3.6 27B (Non-reasoning) | 32.81 |
| 12 | Qwen 3.5 35B A3B (Non-reasoning) | 33 |
| 13 | Claude Opus 4.6 | 33.18 |
| 14 | Qwen 3 VL 32B Instruct | 35.03 |
| 15 | GPT-5 Nano (Minimal) | 35.67 |
Interactive version: theaggregate.ai/benchmark?slug=patterneval-response-pattern-failure-rate-non-thinking · How It Works · Data refreshed daily, snapshot 2026-09-29.