LLM Reference
activeAgents

OSWorld-Verified

Metric: Accuracy (higher is better)Introduced: 2026

Verified OSWorld subset for evaluating desktop computer-use agents on reproducible operating-system tasks. High benchmark score alone doesn't make a model the right pick — weigh it against pricing, API availability, and release date.

Models ranked

8

tracked on this benchmark

Score band

85.0 – 76.7

best → lowest tracked

Snapshot trend

-2.20

May 28 → Jun 30 · 1 models

Leaderboard

Tracked models ranked by Accuracy (higher is better).

Compare candidates
#Model variant and provenanceScore
1
Claude Fable 5
Version: OSWorld-VerifiedHarness: Not recordedEvaluator: Not recordedObserved: Jun 9, 2026Confidence: Not recordedSource

Notes: DAT-5842: Unstarred in Anthropic's shared launch table; starred rows in that table are Mythos 5 only and are intentionally not seeded for Fable 5.

85.0
2
Claude Opus 4.8
Version: OSWorld-VerifiedHarness: Not recordedEvaluator: Not recordedObserved: May 28, 2026Confidence: Not recordedSource

Notes: Official Anthropic launch benchmark; Claude Opus 4.8 exposes computer use as a developer API feature.

83.4
3
Claude Sonnet 5
Version: OSWorld-Verified; 361 tasks, 1080p Ubuntu VM, 100 action steps, adaptive thinking at max effort, pass@1 averaged over 5 runsHarness: Not recordedEvaluator: Not recordedObserved: Jun 30, 2026Confidence: Not recordedSource
81.2
4
Holo3-122B-A10B
Version: OSWorld-Verified - desktop computer useHarness: Not recordedEvaluator: Not recordedObserved: Mar 31, 2026Confidence: Not recordedSource

Notes: DAT-5701: H Company (self-reported); medium confidence; yes - SOTA at March 2026 launch per H Company.

78.8
5
GPT-5.5
Version: OSWorld-VerifiedHarness: Not recordedEvaluator: Not recordedObserved: Jun 9, 2026Confidence: Not recordedSource

Notes: DAT-6106: GPT-5.5 score from Anthropic's Fable 5 comparison table; use as same-harness comparison evidence, not OpenAI self-report.

78.7
6
GPT-5.5 Pro
Version: OSWorld-VerifiedHarness: Not recordedEvaluator: Not recordedObserved: Apr 24, 2026Confidence: Not recordedSource

Notes: Comparable OSWorld-Verified score, but GPT-5.5 Pro does not expose a computer-use API feature in this research pass; this is an eval-harness result.

78.7
7
Holo3-35B-A3B
Version: OSWorld-Verified - desktop computer useHarness: Not recordedEvaluator: Not recordedObserved: Mar 31, 2026Confidence: Not recordedSource

Notes: DAT-5701: H Company (self-reported); medium confidence; yes.

77.8
8
Holotron-3-Nano
Version: OSWorld-Verified - desktop computer useHarness: Not recordedEvaluator: Not recordedObserved: Apr 28, 2026Confidence: Not recordedSource

Notes: DAT-5701: H Company (self-reported); medium confidence; yes.

76.7

How to read this benchmark

This benchmark scores models where higher is better. Scores are useful for directional filtering and shortlisting — not for universal quality ranking. Prefer benchmarks closest to your workload, then validate the linked model pages for pricing, context window, and provider availability.

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.

FAQ

What does the OSWorld-Verified benchmark measure?

Verified OSWorld subset for evaluating desktop computer-use agents on reproducible operating-system tasks. On this page it lists 8 tracked model variants where higher is better.

Is a higher OSWorld-Verified score always better?

For this benchmark, higher is better. A high score helps you shortlist, but confirm pricing, context window, and provider availability on each model page before committing — the top scorer is not always the right pick for your workload or budget.

How current is this OSWorld-Verified data?

This benchmark was last reviewed on Jun 7, 2026. The tracked score average moved -2.20 points across the last 3 snapshots.

Related benchmarks

Last reviewed: Jun 7, 2026

Resources