benchmark evidence
OSWorld-Verified
OSWorld-Verified official computer-use leaderboard.
activeupstream source ->
winner on OSWorld-Verified
direct benchmark result, not a broad vertical composite | source row dated 2026-03-08
scored on 2026-03-08 · stale source data (83d)
latest mapped results | top 20
| # | Model | Score | Evidence | Tested |
|---|---|---|---|---|
| 1 | Anthropic: Claude Sonnet 4.6 | 72.1 | model-only independent_benchmark | 2026-03-08 |
| 2 | Anthropic: Claude Sonnet 4.5 | 62.9 | model-only independent_benchmark | 2025-10-31 |
| 3 | OpenAI: o3 | 23.0 | model-only independent_benchmark | 2025-07-28 |
what this result means
OSWorld-Verified official computer-use leaderboard.
This benchmark contributes direct public evidence. Read its scope before generalizing the result.
A win here is a win on OSWorld-Verified. Broad task pages require independent corroboration before naming a general winner.
source record
category: computer_use
metric: accuracy
matched models: 3
latest source date: 2026-03-08
direction: higher is better