OpenAI has rolled out GPT‑5 to every ChatGPT user, promising sharper reasoning, stronger coding chops, and safer answers while the company hunts a $500 billion valuation. This comes after months of hype and since GPT-4's release back in 2023.
OpenAI says GPT‑5 is “the first time talking to ChatGPT feels like chatting with a real expert.” Sam Altman compares the jump to the first Retina‑display iPhone—after a day with it, he “never wants to go back.”
Key Points
- Big drop today: GPT‑5 now available to free, Plus, Pro, Team, Enterprise users
- Better at coding, math, health, writing; hallucinations cut by nearly half
- Built-in “reasoning” paths and adaptive modes based on task complexity
- Boosts OpenAI’s push toward AGI—though full general intelligence still ahead
The livestream title "LIVE5TREAM" wasn't exactly subtle, but after months of teases and a deleted GitHub leak this morning, GPT-5 is officially here. And unlike the confusing mess of model names OpenAI's been throwing at us lately (o3, o4-mini, o3-mini-high—seriously?), this one's refreshingly straightforward: it's just GPT-5.
GPT‑5 folds three models and a real‑time router into one chat experience. Easy questions get an instant reply; tougher ones trigger “test‑time compute,” letting the AI think longer before answering. When users hit daily caps, a gpt‑5‑mini or gpt-5-nano model takes over so the chat never stalls.
Under the hood, the system shines at code. In demos GPT‑5 spun up a fully working French‑learning web app in seconds and topped coding benchmarks like SWE‑Bench and Aider Polyglot. Writing is much improved: GPT‑5’s prose is tighter and less sycophantic, and the model supports a 400k token context window.
Another thing that's new is how well the model handles uncertainty. Rather than confidently hallucinating when it doesn't know something, GPT-5 uses what OpenAI calls "safe completions"—basically giving you high-level helpful answers that won't send you down a rabbit hole of made-up facts.
GPT-5 will be the default model for free users, while Plus, Team, and Pro plans get higher usage limits; Pro also unlocks GPT‑5 Pro with longer reasoning time. API pricing starts at $1.25 per million input tokens, with cheaper mini and nano options for cost‑sensitive projects.