What Is GPT-5? OpenAI's Self-Routing System and Its Rocky Launch

GPT-5 is the model OpenAI announced on 7 August 2025 and made available the same day to every ChatGPT user, free users included. If you want a one-line answer to "what is GPT-5", it is not a single model but a "system" that decides, request by request, whether a fast model or a deeper-thinking model answers. It merged the GPT line that ran through GPT-4 with the o-series of reasoning models begun in 2024 with OpenAI o1. It is also remembered for the turbulence of its first few days.

A plan to retire the model picker

GPT-5's shape had been announced six months earlier. On 12 February 2025 Sam Altman wrote on X: "We hate the model picker as much as you do and want to return to magic unified intelligence" (as quoted by PCWorld). ChatGPT at the time asked users to choose for themselves among models with different strengths, such as GPT-4o, o1 and o3-mini. Altman said OpenAI would next ship GPT-4.5 as its "last non-chain-of-thought model", and that GPT-5 would then integrate a lot of its technology, including o3. Free users would get unlimited access to GPT-5 at a standard intelligence setting, with Plus and Pro subscribers running it at higher levels.

7 August 2025: a "unified system"

According to OpenAI's announcement, GPT-5 is made of three parts.

The router, OpenAI said, is continuously trained on real signals, including when users switch models, preference rates for responses and measured correctness. Once usage limits are reached, a mini version of each model handles the remaining queries. OpenAI added that it planned to integrate these capabilities into a single model in the near future.

The system card maps the component models onto their predecessors: the fast model (gpt-5-main) succeeds GPT-4o, and the thinking model (gpt-5-thinking) succeeds o3. GPT-5 was available to all users, with more usage for Plus subscribers and GPT-5 pro, with extended reasoning, for Pro subscribers. It launched in the API the same day.

Claimed results and safeguards

OpenAI called GPT-5 "our best AI system yet" and claimed state-of-the-art performance in maths, coding, visual perception and health. All figures are the company's own.

OpenAI also said GPT-5 with thinking outperformed o3 while using 50–80% fewer output tokens, and that on an internal benchmark of knowledge work across more than 40 occupations, including law, logistics, sales and engineering, GPT-5 with reasoning was comparable to or better than experts in roughly half the cases. It emphasised reduced hallucinations and less sycophancy.

On safety, the system card says OpenAI decided to treat gpt-5-thinking as High capability in the biological and chemical domain under its Preparedness Framework, activating the associated safeguards. It did not have definitive evidence that the model could meaningfully help a novice create severe biological harm — its threshold for High capability — but chose a precautionary approach. It was the same kind of call Anthropic had made three months earlier in shipping Claude Opus 4 under ASL-3 as a precaution.

Botched charts, and users who missed GPT-4o

The launch did not go to plan.

First, the bar charts in the livestream were wrong. According to Tech.co, mislabelled bars made the model look better at avoiding deception than it was. Altman called it a "mega chart screwup" on X, and OpenAI's official account apologised for the "unintentional chart crime." On Reddit Altman said the numbers were accurate but the livestream's bar charts had been botched, and that on another slide the numbers were wrong too, blaming human error by tired people working late. The blog's charts were corrected.

The larger problem was the older models. When GPT-5 became the default, GPT-4o, o3 and the other older models disappeared from consumer ChatGPT. Simon Willison noted there was no deprecation period at all (the API still offered the old models). According to Tom's Guide, nearly 5,000 Reddit users weighed in on a thread where some called GPT-5 a "downgrade"; complaints centred on shorter responses, a more formal tone and stricter message caps. The Verge reported grief from users who had treated GPT-4o as a friend or companion.

OpenAI moved quickly. According to The Verge, the day after GPT-5 replaced it, Altman wrote on X: "We will let Plus users choose to continue to use 4o," adding that OpenAI would watch usage as it thought about how long to offer legacy models. Reasoning rate limits for Plus users were raised too. Altman later wrote on Reddit that 4o was back in the model picker for all paid users by default and that, if it were ever deprecated, users would get plenty of notice (per Willison).

GPT-4o's retirement, and the GPTs that followed

The episode shaped the models that came next. With GPT-5.1 on 12 November 2025 OpenAI said it had made ChatGPT's most-used model "warmer", writing that it had heard clearly from users that great AI should be not only smart but enjoyable to talk to.

On 29 January 2026 OpenAI announced that it would retire GPT-4o from ChatGPT on 13 February. The announcement recalled that it had deprecated GPT-4o once and restored it during the GPT-5 release, and said that feedback from a subset of Plus and Pro users who preferred GPT-4o's conversational style and "warmth" had directly shaped GPT-5.1 and GPT-5.2. The reason for retiring it now: most usage had shifted to GPT-5.2, and only 0.1% of users still chose GPT-4o each day. There were no API changes at the time.

The GPT-5 line continued with GPT-5.5 on 23 April 2026. The first GPT-6 models went to paying customers in September 2026, and on 7 October a new GPT-6 model reached all ChatGPT users. The dates of each model are tabulated in the ChatGPT article.

What remained

The point of GPT-5 lay less in its benchmark numbers than in a design shift: users no longer had to think about which model they were using. Its launch also showed that better performance alone does not satisfy users — a familiar tone and the ability to choose are part of the product too. OpenAI itself acknowledged, in retiring GPT-4o, that this feedback had shaped its next models. The wider story is in the AI timeline.