PerceptionBench - Fine-Grained Recognition — leaderboard
Metric: Fine-Grained Recognition (%). Source: huggingface.co. 16 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Seed 2.1 Pro | 56.6 |
| 2 | GPT-5.6 Sol | 55.9 |
| 3 | Kimi K3 | 55.9 |
| 4 | Gemini 3.1 Pro (Preview) | 54.8 |
| 5 | Gemini 3.5 Flash | 51.7 |
| 6 | Claude Fable 5 | 51.6 |
| 7 | GPT-5.5 | 47.2 |
| 8 | Qwen 3.7 Plus | 46.8 |
| 9 | Qwen 3.5 397B A17B | 44.8 |
| 10 | Claude Opus 4.8 | 40.7 |
| 11 | Gemma 4 31B | 39 |
| 12 | GLM-5V Turbo | 36.6 |
| 13 | Grok 4.5 | 36.2 |
| 14 | Kimi K2.6 | 34.5 |
| 15 | GLM-4.6V | 29.3 |
Interactive version: theaggregate.ai/benchmark?slug=perceptionbench-fine-grained-recognition · How It Works · Data refreshed daily, snapshot 2026-08-05.