Benchmarks/agents/OSWorld-G

OSWorld-G

OSWorld-G (Grounding) evaluates screenshot grounding accuracy for OS automation tasks.

Statistics

Total Models1

Average Score0.683

Best Score0.683

Std Deviation—

Properties

Categories

agents grounding multimodal vision

Modalityimage

LanguageEN

Max Score100

Verification

Verified0

Self-reported1

StatusUnverified

Parent Benchmark

Last updated March 16, 2026

Join our newsletter and stay up to date with everything AI

There's too much noise in AI, let's filter it for you. Get a curated digest of models, benchmarks, and the analysis that matters, right in your inbox once a week.

No spam, unsubscribe anytime

The AI Benchmarking Hub.

Leaderboards

AI Leaderboards
LLM Leaderboard
Open LLM Leaderboard
Best AI for Coding
Best AI for Math
Best AI for Image Generation
Best AI for Writing

Arenas

All Arenas
Chat Arena
Coding Arena
Image Arena
Video Arena
Audio Arena
Trading Arena
AI Image Generator
AI Photo Editor

Benchmarks

GPQA
MMLU
MMLU-Pro
AIME 2025
MATH
HumanEval
MMMU
LiveCodeBench
IFEval
GSM8K
SWE-Bench Verified

Models

Gemini 3 Pro
Grok-4 Heavy
GPT-5.1
Grok-4
Qwen3-235B-A22B-Thinking
DeepSeek-R1-0528
GLM-4.6
GPT OSS 120B

Resources

Playground
Blog
News
Community
API
Infrastructure

© 2026 llm-stats

About us Privacy policy Terms of service