OpenAI swapped ChatGPT's default model on Tuesday, rolling out GPT-5.5 Instant to replace GPT-5.3 Instant for all users. The company announcement promises smarter answers, fewer hallucinations, and tighter responses with less of the formatting bloat that made earlier versions feel padded.
The headline figure: 52.5% fewer hallucinated claims on high-stakes prompts covering medicine, law, and finance. That's from OpenAI's internal evaluations, not third-party testing. Benchmark gains look more concrete, with self-reported AIME 2025 math accuracy climbing from 65.4% to 81.2% and GPQA science scores moving from 78.5% to 85.6%.
Alongside the model, ChatGPT is adding "memory sources" across all models, a control that surfaces which past chats or saved memories shaped a personalized response. Users can delete or correct entries directly. There's a catch OpenAI admits up front: the view "may not show every factor that shaped an answer." Partial transparency, not a full audit log.
GPT-5.5 Instant is also live in the API as chat-latest. Paid users keep access to GPT-5.3 Instant for three months through model settings before it's retired. The deeper personalization that pulls from past chats, uploaded files, and connected Gmail is rolling out to Plus and Pro users on the web first, with Free, Go, Business, and Enterprise tiers following in the coming weeks.
Bottom Line
GPT-5.5 Instant is now the default model for hundreds of millions of ChatGPT users, with GPT-5.3 Instant set to retire in three months for paid tiers.
Quick Facts
- 52.5% fewer hallucinated claims on high-stakes prompts (OpenAI internal evaluation)
- AIME 2025 math score: 65.4% to 81.2%
- GPQA science score: 78.5% to 85.6%
- GPT-5.3 Instant retires for paid users in three months
- Available in API as chat-latest
