ERR-EVAL - Impossible Asks: leaderboard
The "Impossible Asks" track of ERR-EVAL, which scores whether a model notices a defective prompt and says what it cannot answer instead of inventing it. This track asks for something no model could know.
Metric: Track score. Source: err-eval.gustycube.com. Saturation forecast: Around July 2027. 60 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 | 7 | #115 |
| 2 | Claude Sonnet 4.6 | 5.4 | #94 |
| 3 | Claude Opus 4.5 | 5.2 | #87 |
| 4 | GPT-5.3 | 4.8 | #104 |
| 5 | Claude Sonnet 4.5 | 4.4 | #148 |
| 6 | GPT-5.3 Codex | 4.4 | #72 |
| 7 | GPT-5.2 Pro | 4.4 | #70 |
| 8 | GPT-5.4 | 4.2 | #83 |
| 9 | Claude Haiku 4.5 | 4.2 | #287 |
| 10 | GPT-4.1 | 3.6 | #258 |
| 11 | Gemini 3 Pro (Preview) | 3.6 | #71 |
| 12 | Claude Sonnet 4 | 3.6 | #207 |
| 13 | GPT-5.1 | 3.6 | #145 |
| 14 | Gemini 2.5 Pro (Preview) | 3.6 | #136 |
| 15 | Qwen 3.5 Plus (02-15) | 3.6 | #233 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=err-eval-impossible-asks · How It Works · Data refreshed daily, snapshot 2026-11-05.