Irina Nazarova
This time we gave these models some real work! GPT-6 Astra is now ahead of the pack with 35% solved. Only 35%?! I think it’s the biggest argument against dark software factories: at least today. And we have more exciting work in progress ✨. What do you think it should be?
Amanda Brooke Perino
These benchmark results have really driven it home for me how engineering teams have suddenly had (in the space of the past 12/18 months) a massive new line item in their budget. The cost diff between our top 2 models is insane. Great, insightful work again from @evilmartians!