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

The author asked GPT-6 Astra to make an OpenAI-style Codex Micro ad, and hours later it had found reference images, built a 3D model with Blender MCP, and produced the audio and final video.

A game developer left Astra with a five-paragraph brief, and in 55 minutes it planned, built, tested and delivered a playable browser city game on its own.

After another update, Astra and Fable 5.1 are level on the AA Intelligence Index, though the poster says their own testing suggests Astra should be ahead.

Astra posted 99.9% on ARC-AGI 3 and 98% on Frontier Math, but Anthropic's Fable 5.1 still leads on broad economic capability.

Built on Astra's rewriting ability, Writing Styles takes under 30 seconds to set up and makes ChatGPT's Slack messages and quick docs sound more like the user.

Given the same Porsche 911 GT3 RS, Astra built a precise model while Fable 5.1 produced a blocky toy, for only $18 more in cost.

During a livestream, GPT-6 Astra was used for Blender previs and posed figures made Seedance's motion stiff, so blocks plus image references worked better.

The author asked Astra to model the M1A1, Hellcat, Bradley and Red Alert's Apocalypse Tank, calling the quality extraordinary and noting it runs in-browser.

Artificial Analysis and Zapier released AutomationBench-AA, a 657-task cross-department business workflow benchmark, where GPT-6 Astra (max) scored 68.5%, ahead of Grok 4.6 and GLM-5.3.

A SaaS founder argues Astra's lower token use per task doesn't make it cost-efficient, since its $10/$50 per million token pricing is far above Sol's $4/$20, and calls for measuring dollars per successful task.

GPT-6 Astra ranks first on Blueprint-Bench 2, a benchmark where AI agents draw floorplans from photographs of apartment interiors, and is close to human performance in 3D spatial understanding.

A post argues Astra's real threat to Anthropic is token efficiency rather than benchmark scores, and that this gap could explain why Claude Code limits run out quickly.