GPT-5.5 Instant Quietly Became the Model Most People Now Use
No keynote, no fireworks — just a default swap on 5 May that put a measurably less hallucination-prone model in front of every ChatGPT user.
- 01GPT-5.5 launched on 23 April 2026; GPT-5.5 Instant became ChatGPT's default model on 5 May, replacing GPT-5.3 Instant.
- 02OpenAI claims GPT-5.5 Instant produces 52.5% fewer hallucinated claims than GPT-5.3 Instant on high-stakes medical, legal and financial prompts.
- 03The model leans agentic — coding, web research, data analysis, document and spreadsheet creation, and moving across tools to finish a task — and can reference past chats, files and Gmail for Plus/Pro users.

The model update that touched the most people in May 2026 didn't get a launch event. On 5 May, OpenAI swapped ChatGPT's default from GPT-5.3 Instant to GPT-5.5 Instant. No fanfare — but the default is the product for the overwhelming majority of users, so this is the change most actually felt.
The lineage
GPT-5.5 itself shipped on 23 April 2026, positioned by OpenAI as another step toward a ChatGPT 'super app'. The 'Instant' variant is the fast, default-tier sibling — tuned for the everyday traffic that makes up most ChatGPT usage rather than the hardest reasoning problems.
The number that matters
The headline claim is about reliability, not raw smarts: in OpenAI's internal evaluations, GPT-5.5 Instant produced 52.5% fewer hallucinated claims than GPT-5.3 Instant on high-stakes prompts — the medicine, law and finance category where a confident fabrication does real damage. If that holds up in the wild, halving the hallucination rate on exactly the prompts where it's most costly is a bigger practical win than a few points on a coding benchmark.
OpenAI also frames 5.5 as clearer and more concise — answers that 'feel better tailored' — which is the kind of soft quality improvement that's hard to benchmark but easy to notice in daily use.
More agent, less chatbot
GPT-5.5's described strengths read like a job description for an agent: writing and debugging code, researching online, analysing data, creating documents and spreadsheets, operating software, and moving across tools until a task is finished. That's consistent with where the whole frontier is heading — models that do multi-step work rather than just answer questions.
A quieter but useful addition: GPT-5.5 Instant can use its search tool to refer back to past conversations, files, and Gmail for more personalised answers, available to Plus and Pro users on the web. It's a memory-and-context play that makes the assistant stickier — and, depending on your priorities, raises the usual questions about how much of your history the model is reaching into.
The read
GPT-5.5 Instant is a confidence-and-reliability release dressed as a routine default swap. OpenAI didn't try to win the month on headline capability — Anthropic and Google were swinging harder there. Instead it shipped the upgrade that quietly improves billions of everyday interactions. Note one thing it is not: despite persistent chatter, there is no announced 'GPT-5.6' — as of early June that remains a rumour, not a release.
Ask Relay — he reads every question himself and replies personally by email.
