PerceptionBench - Fine-Grained Recognition: leaderboard
PerceptionBench slice scoring discrimination between visually similar categories that differ only in fine detail.
Metric: Fine-Grained Recognition (%). Source: huggingface.co. Saturation forecast: Around December 2027. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Seed 2.1 Pro | 56.6 | #73 |
| 2 | GPT-5.6 Sol | 55.9 | #22 |
| 3 | Kimi K3 | 55.9 | #35 |
| 4 | Gemini 3.1 Pro (Preview) | 54.8 | #64 |
| 5 | Gemini 3.5 Flash | 51.7 | #66 |
| 6 | Claude Fable 5 | 51.6 | #20 |
| 7 | GPT-5.5 | 47.2 | #33 |
| 8 | Qwen 3.7 Plus | 46.8 | #98 |
| 9 | Qwen 3.5 397B A17B | 44.8 | #154 |
| 10 | Claude Opus 4.8 | 40.7 | #39 |
| 11 | Gemma 4 31B | 39 | #212 |
| 12 | GLM-5V Turbo | 36.6 | #172 |
| 13 | Grok 4.5 | 36.2 | #54 |
| 14 | Kimi K2.6 | 34.5 | #113 |
| 15 | GLM-4.6V | 29.3 | #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-fine-grained-recognition · How It Works · Data refreshed daily, snapshot 2026-11-05.