OpenAI launches GPT-5 and the model-name chapter quietly closes

On 7 August 2025, OpenAI launched GPT-5. The release was structurally different from previous OpenAI flagship launches. GPT-5 was a single product surface, not a model line-up. Behind the scenes, the system routed between a fast model, a thinking-mode model, and a heavy-reasoning model based on the prompt and the user's tier. As Sam Altman wrote in a same-day post, the routing was "intended to make most users not have to think about which model they are using". The naming chapter, in this framing, had quietly closed.

Most of the prior GPT-4-class line-up was retired. GPT-4, GPT-4 Turbo, GPT-4o, GPT-4o-mini, GPT-4.1 and GPT-4.1 nano left the consumer ChatGPT product over the following two weeks. The o-line was kept available in the API at higher tiers, primarily for backwards compatibility. By mid-September, ChatGPT users had a single model to choose, with subscription tier governing usage limits rather than model selection.

Benchmark numbers, with caveats

The capability gain over Claude 4 and Gemini 2.5 was, on most benchmarks, in the two-to-five-percentage-point range. As reported by The Verge and Ars Technica in coverage of the launch, the headline numbers continued the field's pattern of single-digit improvements per quarter at the frontier. The more substantively different feature was a unified context window of one million tokens for paying users and a roll-out of a new long-running task feature called Projects.

The model number stopped meaning anything. The router did the work.

By the end of August, the unified-router pattern had been adopted in product UI by Anthropic and Google as well. ChatGPT's model picker, which had become a meme through 2024 and early 2025, was reduced to a tier picker. The version of frontier-AI consumer-product UX that had begun with the original ChatGPT in November 2022 had, by August 2025, completed its first full evolutionary cycle.