ChatGPT Gets Smarter Default: GPT-5.5 Instant Is Here
OpenAI rolls out GPT-5.5 Instant as ChatGPT's new default model with 52% fewer hallucinations, concise answers, and a new memory sources feature.
Quick answer
OpenAI replaced ChatGPT's default model with GPT-5.5 Instant on May 5, 2026. The update cuts hallucinated claims by 52.5% in high-stakes topics like medicine and law, uses about 30% fewer words per response, and introduces a memory sources feature that shows users what personal context informed each answer.
ChatGPT Gets Smarter Default: GPT-5.5 Instant Is Here
OpenAI has replaced ChatGPT’s default model with GPT-5.5 Instant, a significant upgrade that focuses on something users have been asking for since day one: fewer made-up answers.
What Changed
GPT-5.5 Instant replaces GPT-5.3 Instant as the model that powers ChatGPT for all users — free and paid. The headline improvement is accuracy: OpenAI says the new model produces 52.5% fewer hallucinated claims on high-stakes prompts in medicine, law, and finance. On conversations that users had previously flagged for factual errors, inaccurate claims dropped by 37.3%.
The model is also noticeably more concise. Responses use roughly 30% fewer words to make the same point, and OpenAI specifically targeted what it calls “gratuitous emojis” and unnecessary follow-up questions — two of the most common complaints about ChatGPT’s tone.
On benchmarks, GPT-5.5 Instant scores 81.2 on the AIME 2025 maths test, up from 65.4 for the previous version, and 76 on MMMU-Pro, up from 69.2.
Memory Sources: A New Transparency Feature
The most interesting new addition is “memory sources.” When ChatGPT uses personal context — from past conversations, uploaded files, or connected Gmail — it now shows you exactly which memories informed its answer. You can review, correct, or delete individual entries. This feature is currently available to Plus and Pro subscribers on the web, with broader rollout coming in the weeks ahead.
What This Means for You
If you use ChatGPT regularly, this is a meaningful upgrade you don’t need to do anything to get. Your conversations should feel tighter and more accurate, especially if you ask about topics where precision matters — health questions, legal queries, or financial planning.
The memory sources feature is worth exploring if you’re a paid subscriber. Being able to see why ChatGPT answered the way it did makes it easier to trust the output — or catch when it’s drawing on outdated context.
For developers, GPT-5.5 Instant is available via the API as the chat-latest model. GPT-5.3 Instant will remain available for paid API users for three months before being retired.
For more on how the full GPT-5.5 model compares to other AI tools, see our GPT-5.5 launch coverage. If you’re new to AI tools, our getting started guide covers the basics, or subscribe to our newsletter for daily updates.
Frequently asked questions
What is GPT-5.5 Instant?
Is GPT-5.5 Instant free to use?
What are memory sources in ChatGPT?
How is GPT-5.5 Instant different from GPT-5.5?
Want to keep learning?
Explore our guided learning paths or try building something with AI right now.
More from News
Alibaba Connects Qwen AI to Taobao for End-to-End Shopping
Alibaba Connects Qwen AI to Taobao for End-to-End Shopping
Alibaba integrates its Qwen AI assistant with Taobao and Tmall, letting users browse, compare, and buy from 4 billion products through conversation.
OpenAI Launches GPT-5.5-Cyber for Security Teams
OpenAI Launches GPT-5.5-Cyber for Security Teams
OpenAI releases GPT-5.5-Cyber, a specialised AI model with fewer guardrails for vetted cybersecurity defenders protecting critical infrastructure.
Anthropic Teaches Claude Agents to 'Dream' and Self-Improve
Anthropic Teaches Claude Agents to 'Dream' and Self-Improve
Anthropic adds dreaming, outcomes, and multiagent orchestration to Claude Managed Agents, letting AI agents learn from past sessions.
Enjoyed this article?
Subscribe for more AI insights delivered to your inbox every week.