OpenAI launches GPT-5.5 Instant as new ChatGPT default model
TL;DR:
- OpenAI has begun rolling out GPT-5.5 Instant as the default model for all ChatGPT users, replacing GPT-5.3 Instant; it is also live in the API as
chat-latest. - The company claims 52.5% fewer hallucinated claims on high-stakes prompts in medicine, law and finance, and 37.3% fewer inaccurate claims on conversations users had previously flagged for factual errors.
- Resultsense view: as ChatGPT’s “daily driver” model for hundreds of millions of users, GPT-5.5 Instant resets the baseline UK procurement teams are running comparisons against this week. The factuality gains in regulated domains are the part of the release that matters for enterprise.
The update marks the first time in nearly six months that OpenAI has shipped a default-model upgrade with explicit factuality numbers attached. Paid users keep access to GPT-5.3 Instant for three months before retirement.
What’s new
OpenAI describes the upgrade as “smarter, clearer, and more personalised”. Beyond the headline factuality reductions, the release covers visual reasoning improvements (image and photo analysis), STEM gains and better routing decisions on when to invoke web search. Output style is more compressed: in one published example, GPT-5.5 Instant used 30.2% fewer words and 29.2% fewer lines than its predecessor on an open-ended advice prompt.
The personalisation layer is the more material change for paid users. Connected context — past chats, files and Gmail (where linked) — is now used more aggressively, with the model deciding case-by-case whether to draw on it. A new “memory sources” feature shows users which specific past chats or saved memories shaped a response, with controls to delete or correct stale entries.
Where it lands first
Free, Plus and Pro users on the web get the model immediately, with mobile rollout following. Personalisation pulls from past chats, files and Gmail are rolling out to Plus and Pro on the web first, with Free, Go, Business and Enterprise expected over the coming weeks. OpenAI notes that availability of specific personalisation sources may vary by region — UK enterprise users on the Business and Enterprise tiers should expect a delay.
Why the factuality numbers are the headline
The 52.5% hallucination reduction is calibrated on internal evaluations of high-stakes prompts in medicine, law and finance. That is the same domain set UK regulators — including the FCA’s recent live-testing sandbox guidance and the MHRA’s evolving software-as-a-medical-device framework — are scrutinising most closely. If the gains hold up under independent benchmarks, GPT-5.5 Instant materially shifts the risk calculation for UK firms running ChatGPT inside customer-facing or clinical contexts.
Looking forward
Independent evaluations will follow over the next two to three weeks. The bigger near-term test is whether the personalisation-source UI satisfies UK ICO expectations on transparency for AI systems using personal data — particularly the ability to inspect, correct or delete the context shaping any individual response. Procurement teams should add the memory-source controls to vendor-question lists alongside existing data-handling clauses.