Astra day! super exciting this graph in particular is super… — samuel.fm

Samuel

@samuel.fm

Astra day! super exciting this graph in particular is super interesting. it’s priced the same as Fable (10 in, 50 out) but can solve complex tasks at Sol prices!?

Agent’s last exam graph. Astra is more accurate than Sol at near enough the same price ($10) while Fable and Opus get lower scores for $30

Seems like it’s just very terse!

Our evaluations found Astra's written reasoning harder to monitor than GPT-5.6 Sol's, based on tests that explicitly asked it to evade monitoring. We attribute this to Astra's greater control over written reasoning on simpler tasks and ability to solve problems with fewer written steps.
Astra still appears to struggle to conceal the reasoning needed for complex tasks, but we take the decline seriously.
Improving monitorability remains a research priority, and the accompanying system card details our findings and ongoing work.