OpenAI Ships GPT-5.5 Instant as ChatGPT's Default Model
OpenAI shipped GPT-5.5 Instant on May 5 as the new default model for all ChatGPT users, completing a two-day rollout. The upgrade targets three pillars — factuality, anti-jailbreak hardening, and baseline intelligence — and pairs them with a redesigned personalization layer that draws on saved memories, past chats, uploaded files, and connected Gmail accounts. Same-day independent benchmarking placed the free-tier model at a capability level that paid models only reached in late 2025.
What the Source Actually Says
The official @OpenAI launch thread describes GPT-5.5 Instant as the new default for all consumer ChatGPT users, also surfacing as gpt-5.5-chat-latest in the API. OpenAI highlights "significant improvements in factuality, especially in domains where accuracy matters most, like medicine, law, and finance," alongside stronger performance on image uploads, STEM questions, and web search routing. OpenAI product engineer @michpokrass — whose launch tweet Altman retweeted — framed the internal focus as "factuality, crushing hacks, and improving the baseline intelligence," describing the result as "much smarter, significantly less likely to hallucinate, and just more delightful to talk to."
The memory upgrade is a tiered rollout: enhanced personalization ships first to Plus and Pro users on web, while Memory Sources — a transparency feature that shows exactly what context informed each response — rolls out to all consumer plans. ChatGPT can now draw simultaneously from saved memories, past chat history, uploaded files, and connected Gmail accounts. Critically, users retain full control: Memory Sources lets them update, delete, or disconnect any individual context source on demand.
Wharton researcher Ethan Mollick (@emollick) provided rare same-day external validation: GPT-5.5 Instant's free tier scores 86% on the GPQA doctoral-level benchmark — a level that even paid models did not reach until late 2025. Mollick noted that GPQA is "fairly consistent and highly correlated with other measured benchmarks," lending the score credibility beyond a single data point. Altman's post-launch commentary framed the experience as more than additive: "the combination of improvements to speed, intelligence, personality, and great memory/personalization feels like a more-than-sum-of-the-parts thing when it all hits together."
Strategic Take
For teams building on the ChatGPT API, gpt-5.5-chat-latest is the alias to track — it will follow the default going forward. The Memory Sources transparency layer signals a shift from raw personalization toward user-controlled context governance, which matters in regulated verticals. The factuality improvements in medicine, law, and finance deserve direct re-evaluation of use cases previously gated by hallucination risk.
