Vibe Check

Vibe Check: GPT-6 Astra Is a Big Upgrade With Some Bad Habits

OpenAI’s new model impressed us with writing, operating software, and visual design. Anthropic’s Fable still has better instincts for building a product.

This is a free preview of a subscribers-only post.

Astra is worth the fuss. We’re still keeping Fable.

For weeks, OpenAI has been previewing Astra’s math results and explaining why its cybersecurity capabilities required stronger safeguards. Now GPT-6 Astra is out, billed as the company’s most intelligent model yet.

Astra gave us writing, consulting work, and designs we wanted to keep. Dan Shipper is enthusiastic about its computer use—operating software through the interface. But a beautiful first result isn’t always a finished product. Fable 5.1 still did better on the complicated apps we tested. Astra also had a habit of adding copy and interface elements the task didn’t need.

We get into the successes and frustrations in our Vibe Check, including:

The first draft of the review itself, which Astra wrote in one shot and Dan thought I (Katie Parrott) had written.
Mike Taylor’s AI training curriculum, built from employee interviews and close to the course he wanted to teach.
Kieran Klaassen’s cozy island, which came with a guided-breathing exercise he hadn’t asked for.

Read the detailed comparisons and how the team’s findings change what we’d hand to Astra next. We’d try it for writing, consulting work, and visual prototypes. For a complicated product, we’d keep reaching for Fable 5.1.

Read the Vibe Check

Join us tomorrow at 12 p.m. ET for our latest camp, How We’re Working Right Now: Fable 5.1 and GPT-6 Astra. We’ll show how we’re using both models and answer your questions. The camp is for paid subscribers only.

Upgrade to join the camp

Katie Parrott is a staff writer at Every. You can read more of her work in her newsletter. To read more essays like this, subscribe to Every, and follow us on X at @every and on LinkedIn.