The grounding numbers: OSWorld 2.0 72.6%, and 47% faster than GPT-5.6 Sol
GPT-6: Astra scores 72.6% on OSWorld 2.0 computer-use tasks and completes computer-use tasks roughly 47% faster than its predecessor, GPT-5.6 Sol, per Artificial Analysis. Speed is the differentiated claim here more than raw accuracy.
Example prompt
Plan a desktop computer-use task that completes the following objective in a real OS environment. Identify the steps most likely to stall, and propose a recovery action for each:
[Describe the desktop task here]Who should look elsewhere
A quarter of OSWorld 2.0 tasks still fail to complete end to end. Build a human checkpoint into any GUI-automation workflow before it touches anything irreversible (payments, deletions, external sends).
Ready to test the workflow?
Recommended access
Use pay-as-you-go API credits for a controlled evaluation on your own workload and cost profile before committing production routing.