Architectural RedlineBench: leaderboard
Models review architectural PDF drawings to identify 55 scored issues. The net score awards full or partial credit for genuine findings and subtracts points for confidently incorrect findings. Reports the publisher’s mean across each model’s runs as a percentage of the available issue points. This board reports Architectural RedlineBench using Mean Net Issue Score (%).
Metric: Mean Net Issue Score (%). Source: redlinebench.benfeicht.com. Saturation forecast: Around October 2027. 31 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5.5 | 68.8 | #3 |
| 2 | GPT-6 Astra | 65.5 | #6 |
| 3 | Claude Opus 5 | 62.7 | #16 |
| 4 | Claude Fable 5.1 | 62.4 | #10 |
| 5 | GPT-6.1 Sol | 59.1 | #7 |
| 6 | Grok 4.7 | 54.2 | #26 |
| 7 | GPT-5.6 Sol | 49.1 | #22 |
| 8 | Claude Fable 5 | 47.9 | #20 |
| 9 | Gemini 4 Argon | 46.4 | #9 |
| 10 | GPT-5.5 | 45.1 | #33 |
| 11 | Claude Opus 4.6 | 41.8 | #67 |
| 12 | Claude Opus 4.8 | 41.2 | #39 |
| 13 | GPT-5.5 Pro | 41.2 | #23 |
| 14 | GPT-5.4 Pro (xHigh) | 40.9 | #38 (GPT-5.4 Pro) |
| 15 | Claude Opus 4.7 | 36.4 | #51 |
Interactive version: theaggregate.ai/benchmark?slug=architectural-redlinebench · How It Works · Data refreshed daily, snapshot 2026-11-05.