The History of AI

GPT-5

Era 5 · Generative AI, reasoning, and agents (2022–2026) › Frontier models

Lineage
Neurons and networks
Kind
Implementation

Who

OpenAI

What happened

OpenAI releases GPT-5, a system that joins a fast model and a reasoning model and decides for each query which one to use. It reaches every ChatGPT user, free users included.

Why it mattered

It unifies OpenAI's general and reasoning models. Users no longer have to pick a model.

The full story

Before GPT-5, OpenAI maintained two lines: general models, such as GPT-4o, and reasoning models, such as o3, which take time to think before answering. Users had to choose from a menu, and Sam Altman had criticized that complexity.

GPT-5, released on August 7, 2025, solved this with a unified system: a fast model for most questions, a deeper reasoning model for hard ones, and a router that decides in real time which to use, based on the complexity of the query or on whether the user explicitly asks it to think harder. It began reaching free users that same day.

OpenAI presented it as its best system to date. It claimed that its responses were 45 percent less likely to contain a factual error than GPT-4o's, and that it outperformed o3 while using 50 to 80 percent fewer output tokens. Altman described it as a significant step toward artificial general intelligence.

The reception was mixed. MIT Technology Review saw it above all as a refined product, far from the transformative future Altman had promised. The router broke down on launch day, and the retirement of the earlier models drew complaints from users who preferred GPT-4o's tone. Altman acknowledged that the company had underestimated how much users valued it and announced that it would return as an option for subscribers.

Sources

How to cite

“GPT-5.” The History of AI, revised November 13, 2026. https://thehistoryofai.com/en/milestone/gpt-5

This entry was last reviewed on November 13, 2026.

Open in the interactive timeline