GPT-5.5 Instant is now the default ChatGPT model — here's what changed
OpenAI replaced GPT-5.3 Instant with GPT-5.5 Instant as ChatGPT's default model on May 5, 2026, bringing fewer hallucinations and tighter responses.
As of 5 May 2026, if you open ChatGPT and start a conversation, you’re talking to GPT-5.5 Instant. OpenAI quietly swapped out GPT-5.3 Instant as the default model, and while it didn’t come with a splashy event, default-model changes are arguably more significant than headline launches. This is the version most people will use, most of the time, without ever touching the model picker.
So what actually got better?
Fewer hallucinations, more reliable answers
The headline improvement is factual accuracy. According to OpenAI’s internal evaluations, GPT-5.5 Instant produces 52.5% fewer hallucinated claims than GPT-5.3 Instant on high-stakes prompts in areas like medicine, law, and finance. On conversations users had already flagged for factual errors, inaccurate claims dropped by 37.3%.
That’s a meaningful shift. If you’ve ever had ChatGPT confidently give you a wrong answer and present it as fact, this is the problem OpenAI has been chipping away at. It won’t be perfect, but the gap between confident and correct is closing.
The model also handles self-correction better. Where GPT-5.3 Instant would sometimes hit a dead end in a maths problem and call it unsolvable, GPT-5.5 Instant catches the error, revisits the working, and reaches the correct answer. That kind of recovery matters more than raw benchmark scores for everyday use.
Tighter, cleaner responses
The other noticeable change is brevity. GPT-5.5 Instant uses 30.2% fewer words and 29.2% fewer lines than its predecessor. OpenAI describes this as cutting “gratuitous emojis and unnecessary overformatting” while keeping the substance.
If you’ve found recent ChatGPT responses a bit verbose or prone to dressing things up with bullet points that didn’t need to be bullet points, this should feel noticeably cleaner. The goal is answers that feel more like a direct response and less like a formatted report.
Smarter use of context and memory
OpenAI is also rolling out enhanced personalisation alongside this update. ChatGPT can now draw on past conversations, uploaded files, and connected Gmail accounts to tailor responses. This is currently available to Plus and Pro users on the web, with Free, Go, Business, and Enterprise access coming in the following weeks.
Alongside this, a new Memory Sources feature gives you visibility into exactly what context was used to personalise a given response. You’ll be able to see whether a reply was shaped by saved memories or previous chats, and you can delete or correct that context directly. Shared chats won’t expose your memory sources to others.
This is a useful step toward ChatGPT actually knowing you over time, rather than starting fresh every session.
Benchmark scores worth knowing
GPT-5.5 Instant scored 81.2 on the AIME 2025 maths benchmark, up from 65.4 for GPT-5.3 Instant. It also improved on the MMMU-Pro multimodal reasoning benchmark, scoring 76 versus 69.2. These aren’t just abstract numbers: they reflect stronger performance on the kind of multi-step reasoning and image analysis tasks that come up in real use.
What this means depending on your plan
The experience varies by plan, so here’s a quick breakdown:
Free users get up to 10 GPT-5.5 messages every 5 hours, after which chats switch to the mini model until the limit resets.
Plus and Go users get up to 160 messages every 3 hours before the same fallback kicks in. Enhanced personalisation from past chats and connected accounts is also rolling out to Plus users now.
Business and Pro users get unlimited access to GPT-5 models, subject to abuse guardrails. Plus and Business users can also manually select GPT-5.5 Thinking from the model picker, with a limit of 3,000 messages per week.
One thing to note on Canvas: GPT-5.5 Instant doesn’t support Canvas. If you use Canvas to create or edit documents, you’ll need to switch to GPT-5.5 Thinking manually.
For developers
GPT-5.5 Instant is available in the API as chat-latest. If you’re currently building on GPT-5.3 Instant, it will remain accessible through model configuration settings for three months before being retired. Worth planning for now rather than scrambling later.
A note on safety classification
One detail worth flagging from the System Card: GPT-5.5 Instant is the first Instant-tier model OpenAI has classified as “High Capability” in its Cybersecurity and Biological and Chemical Preparedness categories. OpenAI is quick to point out that it’s deployed at low reasoning effort, and even at maximum reasoning effort it still performs below GPT-5.5 Thinking. But the classification reflects that the model is genuinely more capable than its predecessors, and OpenAI has put additional safeguards in place accordingly.
The bottom line
GPT-5.5 Instant is a solid, practical upgrade to the model most people use every day. It won’t rewrite how you think about AI, but it makes the daily experience more reliable, more concise, and more context-aware. The hallucination improvements alone make it worth paying attention to, particularly for anyone using ChatGPT for research or professional tasks where accuracy matters.
The best part is you don’t have to do anything. It’s already your default.