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

Astra has completed its full rollout, and OpenAI has also launched GPT-TV, which appears to be a live streaming platform.

A direct request tagging @OpenAI asking whether Plus subscribers can also use Astra in ChatGPT.

A joking reply that Astra probably spent an hour writing two sentences, along with praise for an account's content.

The author notes the proof used nearly 10,000 agents over 88 hours, 270 million messages and 130 billion output tokens at a cost of millions of dollars, with a Lean formal proof.

Before heading out to pick mushrooms, the author tasked Hermès and Astra with making an Obsidian tutorial video using a cloned voice, fully autonomously, and got a link to the result 50 minutes later.

Miles Brundage says a model "significantly more capable than GPT-6 Astra" already exists shortly after Astra's release, and thinks the industry is moving much too fast.

A German post says OpenAI used an unreleased model stronger than Astra to run up to 10,000 agents in parallel for 88 hours, producing a 165-page Lean-verified proof that still awaits independent review.

A user says handing their old hastily vibe-coded code to GPT-6 Astra to clean up has been a rather nice experience.

developedbyed says a week of looping GPT-6 Astra with a few daily steers reached high-fidelity cheeks, and he plans to keep it self-improving for a month with sponsored stickers on the results.

Astra is now available to Plus, Pro, Business, and Enterprise users in Codex and ChatGPT Work, with a live stream showing it at work.

A post claims an internal model stronger than GPT-6 Astra ran about 10,000 concurrent agents for 88 hours, plus 17 hours of Lean formalization, burning roughly 130 billion output tokens on the Navier-Stokes Millennium Prize problem.

A Japanese user shares an introduction video built from slides generated with GPT-Image 2.5 and produced by GPT-6 Astra.