OpenAI rolls out GPT-5.5 Instant as ChatGPT default model
OpenAI is replacing ChatGPT’s default model with GPT-5.5 Instant, which the company says should reduce hallucinations, improve accuracy and make responses more concise. The update is rolling out globally, and GPT-5.3 will be retired after a three-month window for paid users.
OpenAI is replacing ChatGPT’s default model with GPT-5.5 Instant, a new version the company says should produce fewer hallucinations, tighter answers and more reliable responses on topics like finance, law and healthcare.
The release is an incremental update rather than a major product launch. OpenAI said GPT-5.5 Instant is designed to make ChatGPT feel “smarter and more accurate” in general, while also trimming back on overly long replies, unnecessary follow-up questions and what the company called “gratuitous emojis.”
⚡ New to this?
This matters because ChatGPT is changing the model most users talk to by default. A model is the software version doing the actual generating, and OpenAI says this one should make fewer mistakes on sensitive subjects like health, law and money.
The update also changes how ChatGPT uses your past chats and connected services like Gmail, which affects what context it can use to answer you. OpenAI says users can review or delete that context, or use temporary chat instead.
🦞 OpenClaw angle
If you run self-hosted AI agents, treat model upgrades as behavior changes, not just accuracy updates. Re-test your prompts and guardrails on finance, legal, and healthcare-style queries, because OpenAI is explicitly tuning this release for those domains.
If your agents rely on memory, file context, or email integration, add a visible context-review step before each response so users can see what data was used. Also tighten output formatting rules in your system prompts if you want concise answers, since OpenAI is actively reducing verbosity and follow-up questions in GPT-5.5 Instant.
The model replaces GPT-5.3 Instant, which had been the default option for ChatGPT users. OpenAI ships its models in tiers, and the “Instant” version is the one most users interact with unless they change settings manually. Two weeks ago, the company announced GPT-5.5 Thinking and GPT-5.5 Pro, which are aimed at slower, more analytical work and memory-intensive tasks, respectively.
OpenAI said the focus of GPT-5.5 Instant was accuracy, especially in areas where mistakes can have real consequences. Hallucinations - the term used for AI-generated false or made-up information - have been a problem for ChatGPT since its launch in 2022. According to OpenAI, the new model has made “significant improvements” in factuality.
The company pointed to benchmark results as evidence. On HealthBench, a benchmark for medical questions, GPT-5.5 Instant scored 51.4 out of 100, up from 49.6 for GPT-5.3 Instant. On HealthBench Professional, which OpenAI described as a version for clinical use, the score rose from 32.9 to 38.4.
OpenAI said the model also improved on benchmarks related to law and finance. It was tested against conversations that real users had flagged for factual errors, and OpenAI said inaccuracies were reduced by 37.3%.
The company also said GPT-5.5 Instant can draw more actively from users’ previous chats, saved files and connected Gmail accounts to produce more personally relevant answers. When it does that, OpenAI said it will show users what context was used and give them the option to correct or delete it. Users can also switch to a “temporary chat” if they do not want ChatGPT to use that data.
OpenAI said the goal was not only to make responses more accurate, but also to make them feel less cluttered. The company said replies should be “tighter and more to-the-point without losing substance,” while still keeping the tone that ChatGPT users expect. It added that the model asks fewer unnecessary follow-up questions and avoids excessive formatting.
GPT-5.5 Instant is rolling out globally starting today, and OpenAI said all ChatGPT users should see it within a few days. GPT-5.3 will be retired, though paid subscribers will still be able to access it for three months.
