OpenAI has replaced the base ChatGPT model for all users. The main changes are more accurate responses, less «water», and better personalization based on past conversations.
OpenAI has replaced the base ChatGPT model for all users. The main changes are more accurate responses, less «water», and better personalization based on past conversations.
OpenAI has released GPT-5.5 Instant as the new default ChatGPT model — it replaces GPT-5.3 Instant for all users, including the free plan.
The company claims that the new version produces 52,5% fewer hallucinations on sensitive topics — medicine, law, finance — and 37,3% fewer factual errors in complex conversations that users had previously flagged as problematic.
The model also showed significant gains in benchmarks: AIME 2025 in mathematics from 65,4% to 81,2%, PhD level in natural sciences from 78,5% to 85,6%. In addition to accuracy, OpenAI focused on brevity: the model gives shorter answers without losing the essence, asks fewer unnecessary clarifying questions, and avoids excessive formatting.
Another key change is personalization. GPT-5.5 Instant better uses context from past conversations, uploaded files, and connected Gmail to make responses feel more relevant.
In parallel, OpenAI is launching «memory sources» — a tool that shows the user exactly what data influenced the answer and allows them to delete or adjust it.
For paid users, GPT-5.3 Instant will remain available for another three months through model tuning — after which it will be retired.
Enhanced personalization based on past chats and Gmail is currently available for Plus and Pro on the web, with mobile and broader rollout planned in the coming weeks.
