Benchmarks · reasoning
ZeroBench (main)
ZeroBench's 100 main questions: visual problems built to be impossible for contemporary multimodal models. Official leaderboard, pass@5 (solved in at least one of five tries).
60 models · 60 results · data as of 2026-10-29 · zerobench.github.io ↗
Leaderboard
GPT-6 Astra max
All results
60 rows| # | Model | Effort | ZeroBench (main) | Subscription $ / task | API $ / task |
|---|---|---|---|---|---|
| 1 | GPT-6 Astra | max | 52.0% | $0.048 | $0.56 |
| 2 | Opus 5.5 | max | 43.0% | $0.017 | $1.11 |
| 3 | GPT-6 Sol | max | 41.0% | $0.011 | $0.17 |
| 4 | GPT-5.6 Sol | max | 30.0% | $0.042 | $0.76 |
| 5 | Opus 5 | max | 26.0% | $0.012 | $0.68 |
| 6 | Fable 5 | max | 24.0% | $0.11 | $1.01 |
| 7 | GPT-5.4 | xhigh | 23.0% | $0.30 | $0.30 |
| 8 | GPT-5.5 | xhigh | 23.0% | $0.030 | $0.55 |
| 9 | Gemini 3.5 Flash | xhigh | 22.0% | $0.18 | $0.18 |
| 10 | GPT-5.6 Luna | max | 21.0% | $0.026 | $0.28 |
| 11 | Gemini 3 Pro | high | 20.0% | $0.22 | $0.22 |
| 12 | GPT-6 Luna | max | 20.0% | $0.003 | $0.018 |
| 13 | Gemini 3.1 Pro | high | 19.0% | $0.25 | $0.25 |
| 14 | GPT-5.6 Terra | max | 19.0% | $0.030 | $0.55 |
| 15 | Opus 4.8 | max | 18.0% | $0.014 | $0.82 |
| 16 | Gemini 3.6 Flash | high | 18.0% | $0.004 | $0.14 |
| 17 | GPT-5.2 | medium | 18.0% | $0.063 | $0.063 |
| 18 | Grok 4.6 | xhigh | 17.0% | $0.015 | $0.15 |
| 19 | Opus 4.7 | xhigh | 15.0% | $0.32 | $0.32 |
| 20 | Gemini 3 Flash | high | 15.0% | $0.068 | $0.068 |
| 21 | Grok 4.5 | high | 15.0% | $0.023 | $0.15 |
| 22 | Gemini 3.5 Flash-Lite | high | 14.0% | $0.027 | $0.027 |
| 23 | Sonnet 5 | max | 13.0% | $0.012 | $0.69 |
| 24 | Grok 4.3 | high | 13.0% | $0.045 | $0.045 |
| 25 | Opus 4.6 | max | 12.0% | $0.72 | $0.72 |
| 26 | Sonnet 4.6 | max | 12.0% | $0.35 | $0.35 |
| 27 | GPT-5.4 Mini | xhigh | 11.0% | $0.19 | $0.19 |
| 28 | o4 mini | high | 11.0% | $0.071 | $0.071 |
| 29 | Grok 4.20 | xhigh | 10.0% | $0.11 | $0.11 |
| 30 | Opus 4.5 | high | 9.0% | $0.47 | $0.47 |