OpenAI releases GPT-5.5 Instant, a new default model for ChatGPT

On Monday, OpenAI released a caller instauration exemplary called GPT-5.5 Instant, which volition regenerate GPT-5.3 Instant arsenic the default ChatGPT model. The institution said the exemplary reduces hallucination successful delicate areas specified arsenic law, medicine, and finance, portion maintaining the debased latency of its predecessor.

OpenAI released the latest GPT-5.5 exemplary past month with the institution claiming improvements successful areas similar coding and cognition work.

The caller exemplary besides achieved a people of 81.2 successful the AIME 2025 mathematics test, compared to 65.4 for the older model. It besides outperformed its predecessor connected the MMMU-Pro multimodal reasoning benchmark, with a people of 76.0 vs 69.2.

The merchandise placed a peculiar accent connected discourse management. GPT-5.5 Instant tin usage its hunt instrumentality to notation backmost to past conversations, files, and Gmail to springiness you much personalized answers. This diagnostic volition beryllium disposable to Plus and Pro users connected the web, with plans to rotation it retired to mobile soon. OpenAI said that it plans to widen entree to this diagnostic to Free, Go Business, and endeavor users successful the coming weeks.

With this update, ChatGPT volition salso however representation sources crossed each models to assistance you recognize wherever it generated the answers from. Users tin delete outdated sources oregon close them if the reply was wrong. Crucially, the institution said that if you stock a chat with someone, they won’t beryllium capable to spot the representation sources.

For developers, the GPT-5.5 exemplary volition beryllium disposable done API arsenic ‘chat-latest,’ with 5.3 disposable arsenic an enactment for paid users for lone 3 months.

The institution has faced rebuttal from erstwhile exemplary withdrawal moves. When OpenAI withdrew its GPT-4o model, determination was important backlash from users who related to the model’s “personality.” GPT-4o affirmed users’ choices often and that made them consciousness a transportation to that peculiar model. Users who signed petitions to halt OpenAI from retiring it described the exemplary arsenic their “best friend” oregon “a mirror.” Despite the outcry, GPT-4o was deprecated successful February 2026.