AIster · The agentic age · Launches
OpenAI launches GPT-5, then brings GPT-4o back
ChatGPT stopped asking users to pick a model, and the backlash forced a partial U-turn
Key facts
- Date
- 7 Aug 2025
- Era
- The agentic age
- Interest
- Launches
- Threads
- Model wars
- Released
- 7 August 2025
- Design
- Fast model, reasoning model and a real-time router between them
- Access
- Default for all ChatGPT users, including the free tier
- API price
- $1.25 per million input tokens, $10 per million output tokens
OpenAI released GPT-5 on 7 August 2025. OpenAI presented it as a unified system of three parts: a fast model for most questions, a deeper reasoning model for harder ones, and a real-time router that decides which to use. Users no longer had to choose from a list of models themselves.
What changed for users
GPT-5 became the default model for every ChatGPT user, including the free tier, which TechCrunch noted was the first time free users could use an advanced reasoning model. Plus subscribers ($20 a month) got higher usage limits, and Pro subscribers ($200 a month) got unlimited access and a GPT-5 Pro variant. In the API the base model cost $1.25 per million input tokens and $10 per million output tokens.
OpenAI's published scores included 74.9% on SWE-bench Verified, a test of fixing real software bugs, and, for the GPT-5 pro variant without tools, 88.4% on GPQA Diamond, a set of PhD-level science questions. It also reported a much lower rate of factual errors than its earlier o3 model.
The rollout
The launch did not go smoothly. OpenAI removed older models such as GPT-4o from ChatGPT when GPT-5 arrived, and many users complained that the new model felt colder and gave shorter answers. The router also failed on the first day. In a Reddit question-and-answer session the next day, Sam Altman said “the autoswitcher was out of commission for a chunk” of the day, which made GPT-5 seem “way dumber”. A chart in the launch presentation had bars that did not match their numbers, which Altman called a “mega chart screwup”.
OpenAI responded quickly. It doubled rate limits for Plus subscribers, said it would improve how the router made its decisions, and within days restored GPT-4o as an option. Altman later told reporters: “I think we totally screwed up some things on the rollout.” He added: “We've learned a lesson about what it means to upgrade a product for hundreds of millions of people in one day.”
Why it matters
GPT-5 moved ChatGPT from asking people to pick a model to letting the product decide for them. The reaction to losing GPT-4o also showed how attached many users had become to a particular model's tone and personality, and that retiring a model is a product decision with real consequences for the people who rely on it.
Questions people ask
When did GPT-5 come out?
OpenAI released GPT-5 on 7 August 2025 and made it the default model for all ChatGPT users, including those on the free plan.
Why did OpenAI bring back GPT-4o after GPT-5?
Many users objected when GPT-4o was removed at the GPT-5 launch, saying the new model felt colder and gave shorter answers, and a router fault made GPT-5 seem worse on launch day. Within days OpenAI let Plus subscribers choose GPT-4o again.
Sources
- OpenAI, Introducing GPT-5, 2025-08-07
- TechCrunch, OpenAI's GPT-5 is here, 2025-08-07
- TechCrunch, Sam Altman addresses 'bumpy' GPT-5 rollout, bringing 4o back, and the 'chart crime', 2025-08-08