ZeroBench — leaderboard
"Impossible" visual benchmark for multimodal models. SOTA pass@5 is only 19%, 5/5 reliability just 6%. Designed to expose limits of current vision-language reasoning.
Metric: Score (%). Source: zerobench.github.io. Status: saturation imminent. 60 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.4 (xHigh) | 23 |
| 2 | Claude Fable 5 (Max) | 23 |
| 3 | GPT-5.5 (xHigh) | 22 |
| 4 | Gemini 3.1 Pro (Preview) | 19 |
| 5 | Gemini 3 Pro | 19 |
| 6 | Gemini 3.5 Flash | 19 |
| 7 | Claude Opus 4.8 | 17 |
| 8 | GPT-5.2 (Medium) | 17 |
| 9 | Claude Opus 4.7 (xHigh) | 14 |
| 10 | Gemini 3 Flash | 13 |
| 11 | Claude Sonnet 4.6 | 11 |
| 12 | Claude Opus 4.6 | 11 |
| 13 | Claude Opus 4.5 | 10 |
| 14 | GPT-5.4 Mini | 10 |
| 15 | GPT-5.1 (Medium) | 5 |
Interactive version: theaggregate.ai/benchmark?slug=zerobench · How the rankings work · Data refreshed daily, snapshot 2026-07-22.