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

The author speculates that OpenAI may release at least one new model during Tibo's 28 days of shipping, and hopes GPT-6.1 Astra is among them.

The blogger says OpenAI will likely release a new model within 28 days, guessing it will be an Astra 6.1 or 6.5 update timed near Fable 5.5 and Anthropic's IPO.

A 3D mini-game made with Crayon Pro and GPT-6 Astra where you drift and boost a shopping cart and smack five rivals with pool noodles, playable online.

The author reacts to reports that GPT-6 Astra can solve unpublished theoretical physics and math problems, noting AI's strengths are clearly uneven.

The author expects Astra to become the default baseline for robotics papers, but shows a video of a problem it still cannot solve.

On 50 unpublished research-level physics problems from Harvard, Duke and UTokyo physicists, Astra solved 49 without hints at under $3 per problem; Claude Opus 5.5 solved 46.

xdofai previews work on scalable robotics evaluation, with experiments run on both GPT-6 Astra and Fable 5.1.

A rundown of OpenAI's ten days of updates, including 50% faster GPT-6 Astra and 6.1 Sol, repeated global resets and 500k context, though users still favor Opus 5.5.

The author had Astra build a playable prototype in Codex, then used Tripo to generate 3D assets and rig the character, replacing placeholders to finish the game LUMEN ISLES.

A leak says Astra 6.1 arrives as part of OpenAI's 28 Days of Improvements event, mostly small features and bug fixes rather than major changes.

A user reports that picking the top-tier model in WorkBuddy's China edition appears to randomly route to gpt-6 and deduct credits, though this run landed on k3 instead.

A video shows Argon solving difficult 2D sokoban puzzles in one shot that took GPT-6 Astra multiple attempts, though Argon sometimes walks away steps from the solution before returning later.