OSWorld-Verified
Verified OSWorld subset for evaluating desktop computer-use agents on reproducible operating-system tasks.
Models ranked
14
tracked on this benchmark
Score band
85.0 – 76.7
best → lowest tracked
Snapshot trend
+0.10
Oct 9 → Oct 17 · 2 models
Leaderboard
Tracked models ranked by Accuracy (higher is better).
Notes: Unstarred in Anthropic's shared launch table; starred rows in that table are Mythos 5 only and are intentionally not seeded for Fable 5.
Notes: Official Anthropic launch benchmark; Claude Opus 4.8 exposes computer use as a developer API feature.
Configuration: Nex-N2.5-Pro
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. Footnote 8: computer-use benchmarks run with NexCUA harness (to be open-sourced). Retrieved 2026-10-17.
Configuration: Muse Spark 1.1 (xhigh)
Notes: Muse Spark 1.1 model page benchmark table (raw HTML parsed). Methodology: evaluation report https://ai.meta.com/static-resource/muse-spark-1-1-evaluation-report/ (2026-07-09). All Muse Spark 1.1 results run via Meta Model API at xhigh reasoning effort (report §5.1, p.100-101). Report §5.2.1: mean per-task reward (official OSWorld metric), single attempt per task, no shell access. Note 361-task split excludes Google Drive tasks. Retrieved 2026-10-17.
Notes: Meta HF model card Benchmarks table, xhigh reasoning effort.
Notes: H Company (self-reported); medium confidence; yes - SOTA at March 2026 launch per H Company.
Notes: GPT-5.5 score from Anthropic's Fable 5 comparison table; use as same-harness comparison evidence, not OpenAI self-report.
Notes: Comparable OSWorld-Verified score, but GPT-5.5 Pro does not expose a computer-use API feature; this is an eval-harness result.
Notes: H Company (self-reported); medium confidence; yes.
Notes: H Company (self-reported); medium confidence; yes.
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
Last reviewed: Jun 7, 2026