OpenAI’s new GPT-5.5 Instant makes ChatGPT smarter, with more concise and reliable responses

OpenAI Group PBC is replacing the default model in ChatGPT with the launch of GPT-5.5 Instant, and claims that users will notice less hallucinations when it’s discussing “sensitive topics” such as finance, law and healthcare.

ChatGPT should also feel “smarter and more accurate” in general, OpenAI said. In addition, responses should also feel more concise, with the new model cutting back on what the company says is “gratuitous emojis” in its responses.

The new model replaces GPT-5.3 Instant as the default option, and it’s designed to be an incremental upgrade – there’s no new modes and there was no jaw-dropping media demo. There’s also no mention of what happened to GPT-5.4, or why the company skipped ahead to 5.5.

OpenAI ships its GPT models in tiers. Two weeks ago, it announced the launch of GPT-5.5 Thinking and Pro, which are designed for slower, more analytical responses and for memory-intensive tasks, respectively. However, it’s the Instant version that serves as ChatGPT’s default, and the one most users will interact with, unless they go to the settings to adjust things.

Greater accuracy

The goal with this release appears to have been to cut down on ChatGPT’s hallucinations, which have been a problem for the chatbot ever since it launched in late 2022. OpenAI says users should see less made-up facts in its responses, thanks to “significant improvements” in factuality. The new model achieved superior scores to its predecessor on several benchmarks covering domains where accuracy matters most.

For instance, GPT-5.5 Instant generated a score of 51.4 out of 100 on the HealthBench benchmark, which is designed to test AI model’s responses to medical questions. GPT-5.3 Instant had only scored 49.6, so that’s a slight improvement. On HealthBench Professional, which is a version for clinical use, its score jumped from 32.9 to 38.4 points, OpenAI said.

This is an important upgrade because health is one of the most common themes that users discuss with ChatGPT, and so its responses really need to be more reliable. The model recorded similar improvements in accuracy on benchmarks relating to law and finance, and it was further put through its paces on conversations that real users had flagged for factual errors. On these, inaccuracies were reduced by 37.3%, OpenAI said.

In another update, OpenAI has allowed GPT-5.5 Instant to pull information more actively from user’s previous chats, saved files and their connected Gmail accounts, so it can deliver more personally relevant responses to users. When it does this, it will show users the context it used, and give them the option to correct or delete that context. “You remain in control of what’s in your memory,” the company said. Alternatively, users can interact with ChatGPT through a “temporary chat” to stop the model from going through their data.

When OpenAI launched GPT-5.3 Instant about two months ago, it said one of its goals was to make the chat experience feel less “cringe.” With today’s release, it’s focused on making ChatGPT’s responses feel more concise. As a result, responses should feel “tighter and more to-the-point without losing substance,” while retaining the personal touch and warmth that characterizes the ChatGPT experience.

The company specifically mentioned that ChatGPT will also refrain from using emojis too much. “It can deliver the same information, often with more utility than previous models, while reducing the verbosity and overformatting that can make responses too long,” the company wrote. “It also asks fewer unnecessary follow-up questions and avoids things that can make responses feel cluttered, like gratuitous emojis.”

OpenAI said GPT-5.5 Instant is rolling out globally from today, and that all ChatGPT users should be able to experience it within a few days. The previous model, GPT-5.3, will be retired, though paid subscribers will still be able to access it for the next three months.

Images: OpenAI