AI CAN SOLVE MORE. BUT WHAT SHOULD IT SOLVE?

If you could solve any problem for $200, which one would you pick?

We reported at the weekend that OpenAI's next celestial model Astra was solving long-unsolved mathematical problems at a couple hundred dollars a pop.

And if you're of a scientific persuasion, you probably couldn't wait to explore what new mysteries it might help unravel.

Indeed you might not need to wait, as the mathematician Levent Alpöge – who works with Anthropic – reported reproducing five of the ten results from scratch with Claude Fable within 24 hours.

OpenAI also acknowledged other problems were attempted but not solved. So if a current model could reproduce half the set within 24 hours, the choice of problem may be doing rather more work than the headline suggests.

And of course the $200 doesn't price in the failed attempts, picking the problems, OpenAI staff time pulling the work together, and doing what we all have to do – actually verifying the thing.

So, capable as the model clearly is, the return came in part from the choosing. And as solving gets cheaper, choosing well becomes ever more valuable.

In any case, as is now part of the frontier release cycle, OpenAI can’t yet rule out critical cyber capability - Astra will be thoroughly vetted before the rest of us can have a play.

So while the stars might have been visible for a brief moment, it will be a little longer before we see them again.