GPT-6 Astra lands with 75.2% DeepSWE and desktop computer-use scores
- Source
- x.com
- Date
GPT-6 Astra is our best model yet for computer use, software engineering, and visual understanding to date. Hear from the builders putting it to work:
Enter the AGI Era with GPT-6 Astra. GPT-6 Astra scores 75.2% on DeepSWE v1.1 at xhigh reasoning. Ask Astra to fix a bug and have it: • Reproduce the bug from your report • Trace the cause across files • Compare fixes and side effects • Patch the code causing the failure • Run tests and retry the failing workflow
GPT-6 Astra scores 72.6% on OSWorld 2.0 Offline, which tests desktop tasks without internet access. With computer-use tools, Astra can work in the app itself: navigate menus, enter information, and inspect the result on screen. It can switch to code when the task calls for it.
GPT-6 Astra brings stronger visual judgment to front-end design. Give Astra a sketch, reference, or existing UI, and ask it to: • Turn the reference into a working UI • Refine layout, typography, and spacing • Adjust color and interactions • Use screenshots to guide revisions
Context
OpenAIDevs reported that GPT-6 Astra scores 75.2% on DeepSWE v1.1 with its highest reasoning setting, a coding built around realistic bug-fix workflows: reproducing a reported bug, tracing its cause across files, comparing candidate fixes, patching the code, and rerunning tests. On OSWorld 2.0 Offline, a benchmark for desktop tasks performed without internet access, Astra scored 72.6%.
The account also said Astra can use computer-use tools to operate inside an application directly, navigating menus and checking results on screen, and can switch to writing code mid-task when that is the faster path. OpenAI said it is bringing the model to developers on the API and AWS within days of this post.
- benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
Checking sign-in…
Loading comments…