2025 GPT-5
One Model That Decides How Hard to Think
By Andy Oberlin, Founder, AImpact Nexus · Updated November 16, 2026

OpenAI released GPT-5 on August 7, 2025: a single system that routes each request between a fast model and a deeper reasoning model, and the first time that kind of reasoning reached ChatGPT's free users.
Why it matters
Reasoning models stopped being a separate product you had to choose. One system decided for itself when a question deserved more thought, and put that capability in front of hundreds of millions of everyday users.
The story
On August 7, 2025, OpenAI released GPT-5. Instead of asking people to pick between a quick chat model and a slower reasoning model, GPT-5 combined them: a router looks at each request and decides whether to answer fast or to think longer. OpenAI made it the default in ChatGPT, including for free users with usage limits.
What was going on at the time
Between GPT-4 (March 2023) and GPT-5, OpenAI had split its lineup. GPT-4o (May 2024) was fast and multimodal; the o-series, starting with o1 in September 2024, spent extra time reasoning before answering. Users had to know which one to pick. Competitors were shipping their own reasoning models at the same time, and the race was increasingly about reliability and cost, not just raw capability.
How it worked
OpenAI described GPT-5 as a unified system: a fast default model, a deeper reasoning model, and a real-time router that chooses between them based on the conversation and the kind of question. Developers could control how much reasoning effort the model spent. OpenAI said GPT-5 was better at coding, writing and health questions, made fewer factual mistakes, and was less likely to simply agree with the user.
Why it mattered
GPT-5 made 'thinking' AI ordinary. People who had never heard the word reasoning model were now getting answers that had been worked through step by step. For businesses, the question shifted from which model to use to how to point a capable model at their own data and processes.
How people reacted
The launch was not smooth. A chart in the launch presentation contained errors, some users found GPT-5 less warm than GPT-4o, and OpenAI restored GPT-4o as an option within days. Developers generally praised its coding ability, while others argued the jump felt smaller than the leap from GPT-3.5 to GPT-4.
What it left behind
GPT-5 marked the end of the one-big-model-per-year story. AI products became systems: routers, specialist models and tools working together, measured as much on reliability, cost and safety as on benchmarks. That is the world companies are building on today.
Timeline
- March 14, 2023 OpenAI releases GPT-4.
- May 13, 2024 GPT-4o brings fast, multimodal chat to ChatGPT.
- September 12, 2024 o1-preview introduces models that reason before answering.
- August 7, 2025 GPT-5 launches as a unified system with a built-in router, including for free ChatGPT users.
- August 2025 After user feedback, OpenAI restores GPT-4o as an option alongside GPT-5.
Key players
- OpenAI
- Sam Altman
Did you know?
- GPT-5 came out about two and a half years after GPT-4 (March 2023).
- At launch it replaced the model picker in ChatGPT; after user pushback, OpenAI brought GPT-4o back as an option.
- Developers got it in three sizes: GPT-5, GPT-5 mini and GPT-5 nano.
Questions people ask
When was GPT-5 released?
August 7, 2025.
What was new about it?
It combined a fast model and a deeper reasoning model behind a router that decides, per request, how much thinking to do, so users no longer had to pick a model.
Is GPT-5 artificial general intelligence?
No. OpenAI presented it as a major step in capability and reliability, not as AGI.
Further reading
- OpenAI: Introducing GPT-5
OpenAI's own launch announcement.