OpenAI says GPT-5.5 Instant produces 52.5% fewer hallucinated claims “on high-stakes prompts covering areas like medicine, law, and finance”
- GPT-5.5 Instant is rolling out to all ChatGPT users and to the API. — Enhanced personalization is initially for Plus and Pro on the web.
AxiosMegan Morrone
Context & Ripple Effects
GPT-5.5 Instant follows OpenAI’s March rollout of GPT-5.3 Instant, which emphasized answer accuracy and better-contextualized web-search results. The newer model replaces that default in ChatGPT and is also available through the API.
The reported reduction in hallucinated claims is particularly consequential because OpenAI is positioning the default model for uses touching medicine, law, and finance, while extending enhanced personalization first to paid web subscribers.
First-order effects
ChatGPT users are moved from GPT-5.3 Instant to GPT-5.5 Instant as the default model, with OpenAI claiming fewer unsupported claims on high-stakes prompts.
API customers can use the new model immediately; Plus and Pro web users receive the initial enhanced-personalization feature set.
Second-order effects
Developers using the API can reassess workflows that previously needed more verification or narrower prompting, especially where inaccurate answers carry higher operational risk.
The default-model upgrade raises the practical expectation that consumer AI assistants should improve reliability without requiring users to select a more specialized mode or paid tier.
Third-order effects
If successive default-model releases continue to make measurable reliability gains, differentiation among general-purpose assistants will shift further from raw capability claims toward evidence of performance in consequential workflows.
OpenAI’s pairing of broad default-model access with paid personalization suggests a product structure in which baseline quality is widely distributed while customization remains a premium lever.
The trend: This is one data point in the shift from rapid general-model iteration toward making AI assistants dependable enough for more consequential everyday and developer-facing tasks.
GPT-5.5 Instant clearly performs better than GPT-5.3 Instant in both science and math tests. In PhD-level science, it improves from 78.5% to 85.6%, which shows stronger understanding of complex topics. In competition math, it jumps from 65.4% to 81.2%. [image]
🚨 New @OpenAI GPT Model Alert: GPT-5.5 Instant is rolling out in ChatGPT. Key upgrades ⚡ Smarter + clearer responses 🗣️ Warmer, more natural tone ✂️ Much more concise (finally!) ✅ Stronger factuality in medicine, law & finance 🧠 Better memory & personalization 💰 API [video]
GPT-5.5 Instant really said STEM? I got this. GPQA (PhD Science): 78.5% - 85.6% AIME 2025 (Comp Math): 65.4% - 81.2% Math + science upgrades are getting too real. We're cooked (in a good way) [image]
GPT-5.5 Instant is currently rolling out in @ChatGPTapp. I'd still recommend using Thinking mode instead, but it's an improvement over GPT-5.3 Instant which we were stuck with until now. Especially good for Free tier users, who don't have other options!
OpenAI has made GPT-5.5 Instant the default ChatGPT model, significantly reducing hallucinations and improving accuracy across complex reasoning tasks. The update also brings more natural responses and deeper personalization, signaling continued progress toward reliable, [image]
GPT-5.5 Instant isn't just an upgrade... it's a reliability jump ⚡ Better facts Stronger reasoning More trust in high-stakes domains Medicine. Law. Finance. + real gains in vision & STEM tasks This is what “production-ready AI” starts to look like Are we finally there? [image]
We're also improving memory and personalization. ChatGPT can now better use context from saved memories, past chats, files, and connected Gmail accounts to give more personalized responses. Memory sources show what relevant context was used to personalize a response and allow [vi…
GPT-5.5 Instant is more dependable, with significant improvements in factuality, especially in domains where accuracy matters most, like medicine, law, and finance. It's also stronger across everyday tasks, from analyzing image uploads to answering STEM questions to knowing when …
GPT-5.5 Instant is starting to roll out in ChatGPT. It's a big upgrade, giving you smarter, clearer, and more personalized answers in a warmer, more natural tone. And it's also more concise, which we heard you wanted. We think you'll love chatting with it. [video]
New personalization updates in Chat for Plus and Pro are rolling out today! Chat can use more of your information to personalize and you can also see *how* Chat's using what it knows about you via memory sources. Let us know what you think and what you want to see next!