On April 23, OpenAI released GPT-5.5, its most powerful model yet, built for professional users doing complex coding, research, and multi-step work. That model was not for everyone. On May 5, OpenAI started the rollout of GPT-5.5 Instant, replacing GPT-5.3 Instant as the default version of ChatGPT for all users, free or paid, with no settings change required.
Think of it as the everyday edition of OpenAI's frontier model. The full GPT-5.5 was built for heavy-duty professional tasks. GPT-5.5 Instant brings the same generation's improvements to the ChatGPT most people actually use every day for quick questions, writing help, meal planning, research, and everything in between.
Here are 7 things that changed.
Damilare Odedina
1. It makes up far fewer facts, especially on questions that matter most
The biggest upgrade is accuracy. In OpenAI's internal testing, GPT-5.5 Instant produced 52.5% fewer made-up claims than its predecessor on questions about medicine, law, and finance. In real conversations that users had previously flagged for factual errors, wrong claims dropped by 37.3%. For a model used by hundreds of millions of people daily, including many who ask it health and legal questions, those are not small numbers.
2. It is genuinely better at maths, science, and reading images
The accuracy improvements are not just about reducing errors in text. GPT-5.5 Instant scored 81.2 on the AIME 2025 math competition benchmark, up from 65.4 for the previous default. On a PhD-level science reasoning test, it climbed from 78.5 to 85.6. On a test for understanding charts and graphs, it rose from 75.0 to 81.6. If you use ChatGPT to help understand a report, analyse a photo, or work through a maths problem, this model is noticeably stronger at all three.
3. Responses are now 30% shorter. OpenAI built that in deliberately.
GPT-5.5 Instant uses 30.2% fewer words and 29.2% fewer lines than the previous default. OpenAI specifically engineered out the over-explaining, the unnecessary follow-up questions, and the gratuitous emoji that made responses feel cluttered. When you use ChatGPT now, OpenAI says the responses will be tighter and more to-the-point without losing substance
4. ChatGPT will now read your Gmail and past chats to personalise answers
GPT-5.5 Instant can pull context from your past conversations, uploaded files, and your connected Gmail account to make answers personally relevant. Instead of treating every conversation as if it is your first, it remembers your preferences, ongoing projects, and context, so you stop having to repeat yourself. This feature is rolling out now for Plus and Pro subscribers on the web, with Free users coming in the following weeks.
5. A new feature shows you exactly why ChatGPT said what it said
OpenAI is rolling out "memory sources" across all ChatGPT models. When the model uses your past chats or saved memories to shape a response, it now shows you which context it pulled and lets you delete or correct anything that is outdated. If you share a chat with someone else, they cannot see your memory sources. Temporary chats still opt out entirely.
6. This is the most capable everyday ChatGPT model OpenAI has ever shipped
GPT-5.5 Instant is the first model in ChatGPT's standard, everyday tier that OpenAI has applied the same safety protocols previously reserved for its most powerful research-grade models. That is not a warning for regular users. It is a signal of how much more capable this default has become compared to every previous default model. The limitations come with it because the model is powerful enough to warrant them.
7. Paid users have three months before the old model is gone
GPT-5.3 Instant, the previous default, will remain available to paid subscribers for three months through model settings before it is retired. For developers, GPT-5.5 Instant is already live in the API under the alias "chat-latest." If you build products on ChatGPT's API, it is worth testing your prompts now because response length, tone, and structure have all shifted noticeably.
Damilare Odedina
