OpenAI Rolls Out GPT-5.5 Instant as ChatGPT's New Default Model
Original: OpenAI Rolls Out GPT-5.5 Instant as ChatGPT's New Default Model View original →
Overview
On May 5, 2026, OpenAI launched GPT-5.5 Instant as the new default model for all ChatGPT users, replacing GPT-5.3 Instant. Available in the API as 'gpt-5.5-chat-latest', the rollout spans two days globally. Enhanced personalization features are available first for Plus and Pro users on the web.
Key Improvements
- Fewer hallucinations: 52.5% fewer hallucinated claims than GPT-5.3 Instant on high-stakes prompts in medicine, law, and finance
- Better accuracy: 37.3% reduction in inaccurate claims on challenging conversations users flagged for errors
- More concise: 30.2% fewer words and 29.2% fewer lines per response
- Smarter personalization: draws on past chats, uploaded files, and Gmail when connected
Memory Sources Feature
OpenAI is introducing memory sources across all ChatGPT models. When a response is personalized, users can see exactly what context was used — saved memories or past conversations — and delete or correct anything outdated. This transparency feature gives users direct control over how ChatGPT tailors its answers.
Benchmark Performance
GPT-5.5 Instant scored 81.2 on the AIME 2025 math benchmark vs. 65.4 for GPT-5.3 Instant. On the MMMU-Pro multimodal reasoning benchmark, it scored 76 (vs. 69.2 previously).
Legacy Model Timeline
Paid users can continue using GPT-5.3 Instant through model configuration settings for three months before it is retired. The API alias 'chat-latest' now points to GPT-5.5 Instant.
Related Articles
OpenAI launched GPT-5.5 Instant as ChatGPT's new default model, replacing GPT-5.3 Instant. The update delivers 52.5% fewer hallucinations on high-stakes topics like medicine, law, and finance, along with more concise responses and enhanced personalization using Gmail and past conversations.
GPT-5.5 Instant is now ChatGPT's default model, replacing GPT-5.3 Instant. The update cuts hallucinated claims by 52.5% on high-stakes medical, legal, and financial prompts, and adds Gmail-based personalization and memory-source transparency.
The Reddit thread did not stop at “GPT-5.5 is strong.” Its headline mixed together two different official results, and commenters immediately locked onto the harder question: how cheap and repeatable frontier-model cyber capability is becoming.
Comments (0)
No comments yet. Be the first to comment!