Benchmark

ScreenSpot

ScreenSpot is the first realistic GUI grounding benchmark that encompasses mobile, desktop, and web environments. The dataset comprises over 1,200…

Modality
multimodal
Categories
multimodal, spatial_reasoning, grounding, vision
Openness
unknown
Source
llm_stats
Reported scores
16

Reported scores

llm_stats

ModelOrganizationReported valueReportedEvidence
Nova 2 LiteAmazon0.8332025-12-02evidence
Nova 2 OmniAmazon0.8542025-12-02evidence
Nova 2 ProAmazon0.8812025-12-02evidence
Qwen2.5 VL 32B InstructQwen0.8852025-02-28evidence
Qwen2.5 VL 72B InstructQwen0.8712025-01-26evidence
Qwen2.5 VL 7B InstructQwen0.8472025-01-26evidence
Qwen3 VL 235B A22B InstructQwen0.9542025-09-22evidence
Qwen3 VL 235B A22B ThinkingQwen0.9542025-09-22evidence
Qwen3 VL 30B A3B InstructQwen0.9472025-09-22evidence
Qwen3 VL 30B A3B ThinkingQwen0.9472025-09-22evidence
Qwen3 VL 32B InstructQwen0.9582025-09-22evidence
Qwen3 VL 32B ThinkingQwen0.9572025-09-22evidence
Qwen3 VL 4B InstructQwen0.942025-09-22evidence
Qwen3 VL 4B ThinkingQwen0.9292025-09-22evidence
Qwen3 VL 8B InstructQwen0.9442025-09-22evidence
Qwen3 VL 8B ThinkingQwen0.9362025-09-22evidence

Scores are partitioned by the source that reported them and are never merged into a single cross-source ranking, because the sources measure different things and say so.