DISBench (2B Retriever): leaderboard
Metric: Set-level answer F1 (%) on all of DISBench's 122 deep image search queries over personal photo histories, ImageSeeker agent with the Qwen3-VL-Embedding-2B retriever, the model returns the set of target images; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.5 (20251101) | 55.5 | #70 |
| 2 | Gemini 3 Pro (Preview) | 47.9 | #64 |
| 3 | Claude Sonnet 4.5 | 39.4 | #138 |
| 4 | GPT-5.2 | 35.1 | #105 |
| 5 | Gemini 3 Flash (Preview) | 33.9 | #78 |
| 6 | GLM-4.6V | 30.4 | #309 |
| 7 | Qwen 3 VL 32B Instruct | 25.4 | #276 |
| 8 | Qwen 3 VL 235B A22B Instruct | 25.3 | #264 |
| 9 | Qwen 3 VL 235B A22B (Thinking) | 22.8 | #228 (Qwen 3 VL 235B A22B) |
| 10 | GPT-4o | 22.2 | #333 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=disbench-2b-retriever · How It Works · Data refreshed daily, snapshot 2026-10-11.