one year on
OpenAI launches GPT-5
The most anticipated AI release since ChatGPT itself: a unified system with a real-time router deciding when to think, free-tier access for everyone, and aggressive API pricing aimed at the whole market at once.
It’s here. OpenAI launches GPT-5 — not a model but a system: a fast responder, a deeper reasoning engine, and a real-time router that decides per-query how hard to think. The fragmented model picker (4o, o3, 4.5, and friends) is gone; GPT-5 is now the single default for every ChatGPT user, including the free tier.
The claims: state-of-the-art across math, coding, health, and multimodal benchmarks, sharply reduced hallucinations, and best-in-world real-code performance. API pricing lands at $1.25 per million input tokens — a number that reads less like a price and more like a territorial claim.
The launch presentation is polished, confident, and — the internet immediately notices — features bar charts whose proportions defy their own numbers. “Chart crime” trends within the hour.
Expectations for this release have been building since GPT-4 shipped twenty-nine months ago. Whether any model could survive them is tonight’s argument in every group chat in the industry. The stock answer arriving by morning: it’s a very good model, and somehow that feels like an anticlimax — which may say more about 2025 than about GPT-5.
The record
Frames it as 'a PhD-level expert in your pocket' and — in the launch's most-quoted flourish — compares its significance to the Manhattan Project, Death Star imagery included.
Notices the presentation's charts don't quite add up — mislabeled bar heights spawn the instant 'chart crime' meme and a rare same-day apology from Altman.
The loudest first-day story isn't capability — it's grief. OpenAI removes GPT-4o and older models overnight, and the backlash from users attached to 4o's personality forces a reversal within days.
One year later — open only if you can handle spoilers
GPT-5 was simultaneously a technical success and the year's biggest expectations-management failure: benchmarks led, but 'is this all?' set the discourse, the router launched broken, and 4o's forced retirement taught every lab that users form attachments to specific models. The 'AGI vibes' era of launches ended here; the infrastructure era began.
The Weekly Replay · free by email
This week, one year ago — every Sunday.
One email each Sunday: the week's replayed AI news, with the one-year-later annotations included. Written like it's breaking — dated like it isn't.
Free · double opt-in · unsubscribe anytime · privacy