Benchmarks / Reasoning
PerceptionBench
Uses the first-party overall atomic visual-perception accuracy. Score: overall atomic visual-perception accuracy. Collected by Jake Boggs' benchmark dashboard.
PerceptionBench
Higher is better · best effort per model
full table below
| # | Model | Effort | Score |
|---|---|---|---|
| 1 | GPT-5.6 Sol | max | 59.7% |
| 2 | Kimi K3 | max | 58.5% |
| 3 | Fable 5 | max | 57.2% |
| 4 | Gemini 3.1 Pro | high | 56.2% |
| 5 | GPT-5.5 | xhigh | 55.8% |
| 6 | Gemini 3.5 Flash | xhigh | 52.0% |
| 7 | Qwen 3.7 Plus | high | 51.1% |
| 8 | Qwen 3.5 397B-A17B | max | 47.5% |
| 9 | Opus 4.8 | max | 47.2% |
| 10 | Kimi K2.6 | max | 42.6% |
| 11 | Grok 4.5 | max | 41.0% |
| 12 | Gemma 4 31B | max | 40.7% |
| 13 | GLM 5 V | high | 39.6% |
| 14 | MiniMax M3 | max | 33.1% |
| 15 | GLM 4.6 V | low | 32.5% |