multimodal benchmark
VisualWebBench
A multimodal benchmark designed to assess the capabilities of multimodal large language models (MLLMs) across web page understanding and grounding tasks. Comprises 7 tasks (captioning, webpage QA, heading OCR, element OCR, element grounding, action prediction, and action grounding) with 1.5K human-curated instances from 139 real websites across 87 sub-domains.
Updated Aug 7, 2026
Published models2
Registry coverage2
MetricScore
EvidenceB
VisualWebBench leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | AM Nova ProAmazon | 79.7% | 100.0% | 2 | C | |
| 2 | AMNova LiteAmazon | 77.7% | 0.0% | 2 | C |
Score distribution
Top published rows on the benchmark's original scale.
VisualWebBench highlights
The top published results on this benchmark's own scale.
What is VisualWebBench?
Definition and scoring fields from the benchmark registry.
A multimodal benchmark designed to assess the capabilities of multimodal large language models (MLLMs) across web page understanding and grounding tasks. Comprises 7 tasks (captioning, webpage QA, heading OCR, element OCR, element grounding, action prediction, and action grounding) with 1.5K human-curated instances from 139 real websites across 87 sub-domains.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- VisualWebBench
- Modality
- multimodal
- Primary category
- multimodal
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- visualwebbench|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about VisualWebBench.
Which model scores highest on VisualWebBench?
Nova Pro is currently ranked first with 79.7%.
What does VisualWebBench measure?
A multimodal benchmark designed to assess the capabilities of multimodal large language models (MLLMs) across web page understanding and grounding tasks. Comprises 7 tasks (captioning, webpage QA, heading OCR, element OCR, element grounding, action prediction, and action grounding) with 1.5K human-curated instances from 139 real websites across 87 sub-domains.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
2 unique published model results are currently shown.
Does this benchmark affect the overall score?
This benchmark is preserved as source-native evidence but is not eligible for the current overall score.