VisSearch-Bench - DeerFlow - Process Score: leaderboard
VisSearch-Bench (Tsinghua University and Ant Group, arXiv 2611.04894v1) gives multimodal models 204 human-written open-web visual search tasks in five search families. Unlike BrowseComp-VL or MM-BrowseComp, where an image mostly seeds a text search, its tasks require keeping several images, object identities and external facts linked over long searches: finding every item of a visually defined set, chaining dependent visual hops and telling near-identical candidates apart. Systems answer the English questions once per task (Pass@1); temperature 0 except GPT-5.5 and Claude Opus 4.8 (provider default); reasoning effort is not reported. Question access: the 204 scored tasks, input images, reference answers and scoring prompts are public on Hugging Face (antins-labs/VisSearch-Bench-v1). DeerFlow open-source deep-research agent framework driving the listed backbone. Process Score gives task-specific partial credit for correct individual results and sub-answers (%).
Metric: Process score (%, task-specific partial credit, 204 tasks). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 (Benchmark Sweep Run: source=VisSearch-Bench; setting=DeerFlow agent framework; English questions; Pass@1; temperature=provider default; effort unreported) | 38.3 | |
| 2 | Qwen 3.5 397B A17B (Benchmark Sweep Run: source=VisSearch-Bench; setting=DeerFlow agent framework; English questions; Pass@1; temperature=0; effort unreported) | 28.37 | |
| 3 | Qwen 3.6 35B A3B (Benchmark Sweep Run: source=VisSearch-Bench; setting=DeerFlow agent framework; English questions; Pass@1; temperature=0; effort unreported) | 24.26 |
Interactive version: theaggregate.ai/benchmark?slug=vissearch-bench-deerflow-process-score · How It Works · Data refreshed daily, snapshot 2026-11-12.