GPT-6 Astra in Practice: Tests, Use Cases, and Feedback

An OpenAI team member says the team is working to make GPT-6 Astra the best model for frontend and agentic coding, and invites users to try it and share feedback.

A Turkish user had Astra, running at xhigh reasoning, generate a Blender 3D animation of 448 panels continuously reassembling, which took 1 hour 40 minutes.

A user asked GPT-6 Astra to design a desk lamp with an interesting adjustment mechanism; the render looked good and the user says they would like to see it in real life.

Scobleizer had an AI read the posts and discussion about OpenAI's Astra on X and compile a report covering the good, the bad and the ugly.

The author shares comparison images of three models, including GPT-6 Astra at medium setting, answering the same task.

Ryan Greenblatt argues Astra's stronger latent reasoning makes no-CoT experiments more important, and blocking reasoning=None only stops outside researchers from assessing risk.

A Japanese researcher presents "Generative AI and the Future of Scientific Research" at the Japanese Psychological Association, citing GPT-6 Astra's arrival as background for the field's acceleration.

A Japanese illustrator uses Astra to build castle 3D models from literature and research papers, then shoots views from the models to generate card illustrations with noticeably better precision.

A user says GPT-6 Astra on light reasoning spent about 6 minutes of usage, or two Plus-tier limits, to break a single Minecraft block, and calls for the 5-hour limit to be removed.

An Arabic post warns that apps built with GPT-6 Astra in two hours should not ship right away, and that spending 20 extra minutes on checks can prevent a breach within days.

The author shares a preview of one of their tests of GPT-6 Astra for 3D game design and says it crushes the task.

Someone used Astra for world simulation, calls the realism a big step, and shares a generated scene of trees.