ZEROBench-Sub — leaderboard
ZEROBench-Sub — the public subquestion subset of ZeroBench, extremely difficult visual reasoning puzzles over images; models answer the decomposed subquestions of each visual task.
Metric: Score (self-reported). Source: benchmarklist.com. Status: saturation imminent. 40 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | O4 Mini | 29.05 |
| 2 | GPT-5 Mini (High) | 27.79 |
| 3 | GPT-4.5 | 27.01 |
| 4 | GPT-5 (Medium) | 26.19 |
| 5 | Gemini 2.5 Pro | 26.04 |
| 6 | O3 | 25.51 |
| 7 | Claude 3.5 Sonnet | 25.5 |
| 8 | Claude Opus 4.1 | 25.3 |
| 9 | Claude Opus 4 | 25.06 |
| 10 | Gemini 2.5 Flash | 24.48 |
| 11 | Llama 4 Maverick | 23.72 |
| 12 | Gemini 2.0 Flash | 23.24 |
| 13 | Claude Sonnet 4 | 23.01 |
| 14 | O1 Pro | 22.4 |
| 15 | GPT-5 Nano (High) | 21.65 |
Interactive version: theaggregate.ai/benchmark?slug=zerobench-sub · How the rankings work · Data refreshed daily, snapshot 2026-07-22.