PerceptionBench - Localization: leaderboard
PerceptionBench slice scoring where in the image a referenced object actually is.
Metric: Localization (%). Source: huggingface.co. Saturation forecast: Around February 2027. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Sol | 76.7 | #22 |
| 2 | Claude Fable 5 | 70.4 | #20 |
| 3 | Kimi K3 | 70.3 | #35 |
| 4 | GPT-5.5 | 65.8 | #33 |
| 5 | Claude Opus 4.8 | 58.8 | #39 |
| 6 | Gemini 3.1 Pro (Preview) | 52.7 | #64 |
| 7 | Qwen 3.7 Plus | 52.7 | #98 |
| 8 | Gemini 3.5 Flash | 50.6 | #66 |
| 9 | Seed 2.1 Pro | 50 | #73 |
| 10 | Qwen 3.5 397B A17B | 46.7 | #154 |
| 11 | Kimi K2.6 | 45.2 | #113 |
| 12 | Gemma 4 31B | 44.9 | #212 |
| 13 | GLM-5V Turbo | 43.9 | #172 |
| 14 | Grok 4.5 | 39.7 | #54 |
| 15 | MiniMax-M3 | 33.3 | #151 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=perceptionbench-localization · How It Works · Data refreshed daily, snapshot 2026-11-05.