VisSearch-Bench - Direct Answer - 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). Direct Answer: the model answers from the question and its images alone, with no tools. 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. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 (Benchmark Sweep Run: source=VisSearch-Bench; setting=Direct Answer, no tools; English questions; Pass@1; temperature=provider default; effort unreported) | 22.61 | |
| 2 | Gemini 3.5 Flash (Benchmark Sweep Run: source=VisSearch-Bench; setting=Direct Answer, no tools; English questions; Pass@1; temperature=0; effort unreported) | 22.28 | |
| 3 | Claude Opus 4.8 (Benchmark Sweep Run: source=VisSearch-Bench; setting=Direct Answer, no tools; English questions; Pass@1; temperature=provider default; effort unreported) | 16.13 | |
| 4 | DeepSeek V4.1 Flash (Benchmark Sweep Run: source=VisSearch-Bench; setting=Direct Answer, no tools; English questions; Pass@1; temperature=0; effort unreported; checkpoint date unstated) | 12.88 | |
| 5 | Qwen 3.5 397B A17B (Benchmark Sweep Run: source=VisSearch-Bench; setting=Direct Answer, no tools; English questions; Pass@1; temperature=0; effort unreported) | 12.72 | |
| 6 | Qwen3.8-Flash-Next (Benchmark Sweep Run: source=VisSearch-Bench; setting=Direct Answer, no tools; English questions; Pass@1; temperature=0; effort unreported; listed as Qwen3.8-Flash in the paper text) | 11.82 | |
| 7 | Qwen 3 VL 235B A22B (Thinking) (Benchmark Sweep Run: source=VisSearch-Bench; setting=Direct Answer, no tools; English questions; Pass@1; temperature=0; effort unreported) | 11.37 | |
| 8 | Kimi K3 (Benchmark Sweep Run: source=VisSearch-Bench; setting=Direct Answer, no tools; English questions; Pass@1; temperature=0; effort unreported) | 10.24 | |
| 9 | MiniMax-M3 (Benchmark Sweep Run: source=VisSearch-Bench; setting=Direct Answer, no tools; English questions; Pass@1; temperature=0; effort unreported) | 9.71 | |
| 10 | Qwen 3.6 35B A3B (Benchmark Sweep Run: source=VisSearch-Bench; setting=Direct Answer, no tools; English questions; Pass@1; temperature=0; effort unreported) | 9.55 |
Interactive version: theaggregate.ai/benchmark?slug=vissearch-bench-direct-answer-process-score · How It Works · Data refreshed daily, snapshot 2026-11-12.