Coding agents completed an estimated 46% of OpenAI’s two-to-four-hour research tasks without human intervention in October, up from 23% in March, according to a new OpenAI report on how AI is speeding up its own research. The median researcher spent the equivalent of $2,400 a day on agents at API list prices in October, a 62% jump from a month earlier. Success rates still vary widely by job, from 79% when watching training runs to 25% when writing inference and serving code. Researchers also credited people with nearly all of the key ideas behind GPT-6 Astra and GPT-6.2 Astra. OpenAI said it published the figures because it thinks labs should have to publicly report signs that AI is accelerating the development of future AI.





