multimodal benchmark
ScreenSpot
ScreenSpot is the first realistic GUI grounding benchmark that encompasses mobile, desktop, and web environments. The dataset comprises over 1,200 instructions from iOS, Android, macOS, Windows and Web environments, along with annotated element types (text and icon/widget), designed to evaluate visual GUI agents' ability to accurately locate screen elements based on natural language instructions.
Updated Aug 7, 2026
Published models16
Registry coverage16
MetricScore
EvidenceB
ScreenSpot leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | AC | 95.8% | 100.0% | 16 | C | |
| 2 | AC | 95.7% | 93.3% | 16 | C | |
| 3 | AC | 95.4% | 86.7% | 16 | C | |
| 4 | AC | 95.4% | 80.0% | 16 | C | |
| 5 | AC | 94.7% | 73.3% | 16 | C | |
| 6 | AC | 94.7% | 66.7% | 16 | C | |
| 7 | AC | 94.4% | 60.0% | 16 | C | |
| 8 | AC | 94.0% | 53.3% | 16 | C | |
| 9 | AC | 93.6% | 46.7% | 16 | C | |
| 10 | AC | 92.9% | 40.0% | 16 | C | |
| 11 | AC | 88.5% | 33.3% | 16 | C | |
| 12 | AM Nova 2 ProAmazon | 88.1% | 26.7% | 16 | C | |
| 13 | AC | 87.1% | 20.0% | 16 | C | |
| 14 | AMNova 2 OmniAmazon | 85.4% | 13.3% | 16 | C | |
| 15 | AC | 84.7% | 6.7% | 16 | C | |
| 16 | AMNova 2 LiteAmazon | 83.3% | 0.0% | 16 | C |
Score distribution
Top published rows on the benchmark's original scale.
ScreenSpot highlights
The top published results on this benchmark's own scale.
What is ScreenSpot?
Definition and scoring fields from the benchmark registry.
ScreenSpot is the first realistic GUI grounding benchmark that encompasses mobile, desktop, and web environments. The dataset comprises over 1,200 instructions from iOS, Android, macOS, Windows and Web environments, along with annotated element types (text and icon/widget), designed to evaluate visual GUI agents' ability to accurately locate screen elements based on natural language instructions.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- ScreenSpot
- Modality
- multimodal
- Primary category
- multimodal
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- screenspot|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about ScreenSpot.
Which model scores highest on ScreenSpot?
Qwen3 VL 32B Instruct is currently ranked first with 95.8%.
What does ScreenSpot measure?
ScreenSpot is the first realistic GUI grounding benchmark that encompasses mobile, desktop, and web environments. The dataset comprises over 1,200 instructions from iOS, Android, macOS, Windows and Web environments, along with annotated element types (text and icon/widget), designed to evaluate visual GUI agents' ability to accurately locate screen elements based on natural language instructions.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
16 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.