This is a free preview of a subscribers-only post.
Was this newsletter forwarded to you? Sign up to get it in your inbox.
GPT-6.2 Sol planned a trip so well that Mike Taylor, Every’s head of evals, gave it his first perfect score on the task. He still doesn’t plan to use it.
OpenAI launched Sol alongside GPT-6.2 Astra, promising an everyday workhorse and a model for your hardest projects. Seven of us tested them for more than two weeks. We found Astra useful for clearly defined coding and operations work, but its longer writing lost the thread and its ambitious builds needed prodding. Sol gave most testers even fewer reasons to switch.
Katie Parrott’s full Vibe Check shows how we reached that verdict, including:
- Mike Taylor’s travel test and why a perfect itinerary didn’t win a place in his daily work.
- Kieran Klaassen’s coding builds, where Astra’s speed came with a problem in the tests it left behind.
- Dan Shipper’s manuscript review, where Astra missed a factual contradiction that Fable caught.
- Tyler Nishida’s interface designs, with inventive ideas, unfinished interactions, and stubborn visual defaults.
Astra is worth trying when you can define the job and check the result. For open-ended writing and building, our team still reaches for Claude and Gemini. The full review includes the benchmarks, screenshots, and individual verdicts that explain where the models fit—and where to save yourself a test.
Katie Parrott is a staff writer at Every. To read more essays like this, subscribe to Every, and follow us on X at @every and on LinkedIn.
|
|
What is included in a subscription?
Daily insights from AI pioneers + early access to powerful AI tools
|
||||||||
|