Enterprise AI in Numbers · Capability and cost
Top score on OSWorld 2.0, the benchmark of agents operating real computer apps
31.4%
31.4% (Claude Opus 5) as of July 24, 2026, according to Epoch AI, Benchmarking hub, up from 27.3% on July 9, 2026. The highest share of OSWorld 2.0 tasks (multi-step work in real desktop and web applications) that any evaluated model completes, with its lab.
What it means for leaders
Agents that operate the same screens people use can automate work in systems without APIs, from back-office tools to online checkout. Scores are still low, so keep these agents on narrow, supervised tasks.
What would change the reading: The first models to complete a majority of tasks.
The trend
Every reading
Definition and method
- What it measures
- The highest share of OSWorld 2.0 tasks (multi-step work in real desktop and web applications) that any evaluated model completes, with its lab.
- Population
- AI models evaluated on OSWorld 2.0, as compiled by Epoch AI
- Updated
- When new models are evaluated
- Source
- Epoch AI, Benchmarking hub · CC BY 4.0
Caveats
- Scores come from independent evaluators compiled by Epoch AI, not from the labs, and depend on the settings and tools each run used.
- The leader is the best result so far, dated by the release of the model that set it; a model evaluated late can take the lead after its release date.
- OSWorld 2.0 started in mid-2026, so its history is short.
Cite this
Enterprise AI in Numbers, "Top score on OSWorld 2.0, the benchmark of agents operating real computer apps", https://lucianoayres.com/ai-brief/data/computer-use-benchmark-leader/, retrieved October 25, 2026. Data: Epoch AI, Benchmarking hub.