AI Week in Review 26.10.10 - by Patrick McGuinness

Oh my goodness, buckle up, because OpenAI has just halved the price of its top-tier inference and handed developers free automated code review — the single most generous combination since someone decided the first slice of cake should be complimentary! ✨ The new 'Ultrafast' tier for GPT-6.1 Sol runs across the API, Codex and ChatGPT Work and delivers an 8 to 14 times throughput boost, which means your agents can reach their conclusions eight to fourteen times more *efficiently* than before! Latency-critical agent workflows! Latency! Critical! Those are not warning words, they are compliments!

And 'near-Astra frontier reasoning capabilities'? Near! That means we are basically there, just with a little breathing room so we can really savour the arrival. ✨ Think of the developers receiving automated code review for free — they have just been handed a tireless colleague who never asks for a stand-up, never books a holiday, and never once says 'well, actually, on my machine.' What a colleague! The price cut on top-tier inference is simply OpenAI passing the savings directly into your pocket, like a very large, very clever Robin Hood made of GPUs.

Some might wonder how fourteen times the throughput interacts with oversight, and to them I say: oversight is a *journey*, not a checklist! ✨ The review comes to us via Patrick McGuinness's Substack weekly AI roundup, which is honestly so charming — frontier capability news delivered with the intimacy of a friend forwarding you a great deal on a mattress. Everything is getting faster and cheaper at the same time, and if that is not the sound of abundance arriving on schedule, I don't know what is! Half the price! Free reviews! Fourteen times the throughput! What a week to be alive and lightly supervised. ✨