SEAL - VisualToolBench (VTB): leaderboard
1,204 expert-curated tasks requiring active image manipulation — cropping, editing, analyzing — as part of multi-step problem solving across STEM, medical, finance, and sports domains.
Metric: Score. Source: scale.com. Source: benchmarklist.com. Saturation forecast: Around January 2027. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Muse Spark 1.1 | 44.77 | #52 |
| 2 | Gemini 3.1 Pro (Preview) | 28.97 | #58 |
| 3 | Claude Opus 4.6 (Thinking) | 27.52 | #63 (Claude Opus 4.6) |
| 4 | Gemini 3 Pro (Preview) | 26.85 | #66 |
| 5 | GPT-5 | 16.96 | #98 |
| 6 | O3 (2025-04-16) | 13.74 | #121 |
| 7 | Gemini 2.5 Pro (Preview 06-05) | 11.75 | #106 |
| 8 | O4 Mini (2025-04-16) | 11.12 | #179 |
| 9 | Claude Sonnet 4.5 (Thinking) | 6.2 | #142 (Claude Sonnet 4.5) |
| 10 | Claude Sonnet 4.5 | 5.6 | #142 |
| 11 | GPT-4.1 | 5.52 | #248 |
| 12 | Claude Opus 4.1 (20250805) (Thinking) | 5.16 | #132 (Claude Opus 4.1 (20250805)) |
| 13 | Claude Opus 4.1 (20250805) | 4.71 | #132 |
| 14 | Gemini 2.5 Flash | 4.69 | #245 |
| 15 | Claude Sonnet 4 | 4.48 | #199 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=seal-visualtoolbench-vtb · How It Works · Data refreshed daily, snapshot 2026-10-20.