OpenAI adjusted the default model behind ChatGPT on May 28, fine-tuning it to produce replies that are easier to read, more natural and better paced. The update also reduced the frequency of bullet-heavy responses, according to the company’s release notes. In the same announcement, OpenAI said it would retire the o3 and GPT-4.5 models from ChatGPT on August 26 and June 27, respectively, while stressing that the move does not affect API access.
The twin signals underscore the company’s push to streamline the ChatGPT model picker around the GPT-5.5 family and deliver a more consistent conversational experience. For API customers relying on the chat-latest alias, the change carries a separate significance: the pointer now leads to a model priced at $5 per million input tokens and $30 per million output tokens, well above the prior generation’s $2.50/$15.
GPT-5.5 Instant first became ChatGPT’s default on May 5, replacing GPT-5.3 Instant and taking over the chat-latest API alias. At launch, OpenAI published internal evaluation data showing the model generated 52.5% fewer hallucinated claims on high-stakes prompts in medicine, law and finance, and 37.3% fewer inaccurate claims in user-flagged conversations that had proved difficult for earlier systems. The model also added the ability to draw on personalization context—past chats, files and, where available, Gmail—with memory-source controls that show users some of that context.
The May 28 update did not alter the model identity but tuned the serving behavior. OpenAI said the adjustments would make answers “easier to read, more natural, better paced” and curb long bullet-heavy replies. The improvements apply to both ChatGPT users and developers hitting the chat-latest endpoint.
Separately, the company announced the retirement of two older ChatGPT models. OpenAI o3 will be removed from the service on August 26, 2026, after a 90-day sunset; GPT-4.5 will be removed on June 27, 2026, after 30 days. The release note explicitly states that these changes affect ChatGPT only and there are no API deprecations for those models.
OpenAI’s API deprecations page lists a different set of sunsets: on May 8 the company declared that gpt-5.2-chat-latest and gpt-5.3-chat-latest snapshots would be shut down on August 10, 2026. Those are unrelated to the ChatGPT retirements. For developers using chat-latest, the practical effect is that their calls now route to GPT-5.5, with its higher token pricing compared with the now-deprecated GPT-5.4 generation.
The precise nature of the May 28 update—whether it involved a new model snapshot or was a runtime configuration change—has not been disclosed, leaving chat-latest users without clarity on behavioral stability. OpenAI has not said whether enterprise or education customers will receive different retirement timelines or legacy access to o3 and GPT-4.5. The company has also not released an updated system card with full benchmark methodology, and no margin or cost data accompanied the update, making any margin impact claims speculative.
For the millions of ChatGPT users, the changes mean a default model that speaks more naturally and a shrinking menu of legacy choices. For API customers, the rise of GPT-5.5 through chat-latest couples an improved model with a higher price point, a shift that will show up in monthly bills as older endpoints fade.