activeAgents

OSWorld-G

Metric: Accuracy (higher is better)Introduced: 2025

OSWorld grounding benchmark variant for measuring GUI element grounding and localization in desktop computer-use environments.

Models ranked

6

tracked on this benchmark

Score band

87.4 – 61.6

best → lowest tracked

Snapshot trend

+9.85

Nov 13 → Oct 17 · 1 models

Leaderboard

Tracked models ranked by Accuracy (higher is better).

Compare candidates

#Model variant and provenanceRelative to leaderScore

1

Configuration: Nex-N2.5-Pro

Version: OSWorld-G (UI grounding)Harness: Grounding eval (0-1000 normalized coords)Evaluator: Nex-AGIObserved: Oct 17, 2026Confidence: self-reportedSource

Notes: Nex-AGI HF model card Text/Multimodal Benchmarks tables (also HF eval-results widget for coding rows). Sampling temperature 0.7, top_p 0.95, top_k 40. Grounding benchmark; static-style localization eval. Retrieved 2026-10-17.

87.4

2

Holo2-235B-A22B

Version: OSWorld-G - UI groundingHarness: Not recordedEvaluator: Not recordedObserved: Nov 13, 2025Confidence: Not recordedSource

Notes: H Company (self-reported); medium confidence; yes.

79.0

3

Version: OSWorld-G - UI groundingHarness: Not recordedEvaluator: Not recordedObserved: Nov 13, 2025Confidence: Not recordedSource

Notes: H Company (self-reported); medium confidence; yes.

76.1

4

Holotron-3-Nano

Version: OSWorld-G - UI groundingHarness: Not recordedEvaluator: Not recordedObserved: Apr 28, 2026Confidence: Not recordedSource

Notes: H Company (self-reported); medium confidence; yes.

72.0

5

Version: OSWorld-G - UI groundingHarness: Not recordedEvaluator: Not recordedObserved: Sep 25, 2025Confidence: Not recordedSource

Notes: H Company (self-reported); medium confidence; yes.

71.8

6

Version: OSWorld-G - UI groundingHarness: Not recordedEvaluator: Not recordedObserved: Sep 25, 2025Confidence: Not recordedSource

Notes: H Company (self-reported); medium confidence; yes.

61.6

How to read this benchmark

This benchmark scores models where higher is better. Use scores for directional filtering and shortlisting, not universal quality ranking; then validate pricing, context window, provider availability, and fit for your workload.

Trust this score when

There is a fresh timestamped snapshot (or multiple snapshots) for this benchmark.
The model list covers the same version family you can actually deploy today.
Top candidates overlap with your required routing and feature requirements.

Be cautious when

There is only one benchmark snapshot or the dataset appears stale.
The benchmark metric direction is opposite of your decision objective.
The score difference between options is narrow and likely within implementation variance.

Related benchmarks

AA Coding Agent Index

Artificial Analysis Coding Agent Index

Agents' Last Exam

AutomationBench

Berkeley Function Calling Leaderboard v3

Last reviewed: Jun 7, 2026

Resources