PerceptionBench - Counting: leaderboard
PerceptionBench slice scoring exact enumeration of objects in a scene.
Metric: Counting (%). Source: huggingface.co. Saturation forecast: Around September 2027. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Sol | 62.4 | #22 |
| 2 | Kimi K3 | 59.7 | #35 |
| 3 | Gemini 3.1 Pro (Preview) | 56.9 | #64 |
| 4 | GPT-5.5 | 55.8 | #33 |
| 5 | Qwen 3.7 Plus | 53.3 | #98 |
| 6 | Claude Fable 5 | 52.9 | #20 |
| 7 | Seed 2.1 Pro | 51.2 | #73 |
| 8 | Qwen 3.5 397B A17B | 49.1 | #154 |
| 9 | Kimi K2.6 | 45.2 | #113 |
| 10 | Claude Opus 4.8 | 44.2 | #39 |
| 11 | Gemini 3.5 Flash | 43.6 | #66 |
| 12 | GLM-5V Turbo | 40 | #172 |
| 13 | Grok 4.5 | 35.2 | #54 |
| 14 | Gemma 4 31B | 33.9 | #212 |
| 15 | GLM-4.6V | 31.8 | #330 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=perceptionbench-counting · How It Works · Data refreshed daily, snapshot 2026-11-05.