GPT-5: OpenAI puts a fast model and a reasoning model behind an automatic router
In one sentence On August 7, 2025 OpenAI releases GPT-5: a unified system where a router picks between a fast model and a deeper reasoning model. The API gets GPT-5, mini and nano; in ChatGPT it becomes the default model, free users included.
GPT-5 changes how ChatGPT is used. Before, users had to pick between models (GPT-4o for quick answers, o3 for long reasoning). GPT-5 is a single system instead: a fast model answers most questions, a reasoning model steps in on hard problems, and a "router" decides in real time which one to use.
In ChatGPT, GPT-5 becomes the default for everyone, free users included (with usage limits). Pro subscribers also get GPT-5 pro, a version that reasons longer. For developers, the API offers three sizes: GPT-5, GPT-5 mini and GPT-5 nano.
The reception was not only positive. The launch abruptly removed GPT-4o and the other models from ChatGPT, and many users protested. Within days Sam Altman admitted the router was not working as intended and made GPT-4o available again to Plus subscribers as a "legacy" model.
Companies
OpenAI
Tools
GPT-5, GPT-5 mini, GPT-5 nano, GPT-5 pro
Tags
Sources
- https://cdn.openai.com/gpt-5-system-card.pdf
- https://developers.openai.com/docs/models/gpt-5
- https://www.vellum.ai/md/blog/gpt-5-benchmarks
- https://redmondmag.com/articles/2025/08/07/microsoft-embeds-chatgpt5-across-copilot-ecosystem.aspx
- https://www.techrepublic.com/article/news-openai-reinstates-gpt4o/