GPT-5.5 Instant Is Now ChatGPT's Default Model: What Changes and Why It Matters
On May 5, 2026, OpenAI updated ChatGPT's default model to GPT-5.5 Instant, replacing GPT-5.3 Instant for all users. This isn't a premium model launch — it's a change to the model that hundreds of millions of people use every day, which makes it one of the highest-impact updates of the year.
Most ChatGPT users never manually switch models. They open the app, type their question, and get a response from whatever the default model is. Changing that default affects an enormous number of conversations daily.
What GPT-5.5 Instant Improves Over Its Predecessor
OpenAI published improvement data in their official announcement:
Reduced hallucinations: GPT-5.5 Instant produces 52.5% fewer incorrect claims than GPT-5.3 Instant on high-stakes prompts — medicine, law, and finance. In conversations that users had flagged for factual errors, inaccurate claims dropped by 37.3%.
More concise and natural responses: the model better matches format to the task. For casual advice questions, it avoids unnecessary bullet points and follow-up questions that made previous responses feel mechanical.
Better context handling: GPT-5.5 Instant manages previously provided information more effectively. It doesn't re-ask questions the user already answered and maintains coherence across long conversations.
Published benchmarks: according to TechCrunch, the model scored 81.2 on the AIME 2025 math test, compared to 65.4 for the previous model. On the MMMU-Pro multimodal reasoning benchmark, it scored 76 versus 69.2 for GPT-5.3 Instant.
Expanded Memory and Personalization
One of the most notable new features is the expansion of memory sources the model can draw from to personalize responses. GPT-5.5 Instant can now access:
- Relevant past conversations
- Files the user has previously shared
- Gmail (for users who have it connected)
The model automatically decides when to activate personalization — it doesn't apply it to every response, only when personal context genuinely adds value.
This feature is initially available for Plus and Pro plan users on the web. Free, Go, Business, and Enterprise plans will receive expanded personalization in the coming weeks.
Availability and Plans
GPT-5.5 Instant is available for all ChatGPT users starting May 5, including the free plan. There is no additional cost — it's the new standard model.
Paid users have the option to keep GPT-5.3 Instant for three more months, accessible through model configuration settings, before it's permanently retired.
For developers, GPT-5.5 Instant is available in the OpenAI API as chat-latest. GPT-5.5 and GPT-5.5 Pro arrived in the API on April 24, 2026. Updated pricing is at openai.com/api/pricing.
Context: OpenAI's Update Pattern
This launch follows the pattern OpenAI has established in recent months: frequent updates to the default model with incremental improvements in factuality and usability, rather than waiting for entirely new model releases.
GPT-5.3 Instant had replaced GPT-5.1 Instant in February 2026, when OpenAI also retired the original GPT-5 models. GPT-4o, which generated significant backlash when deprecated that same month — users described it as their "best friend" — is no longer available.
The update cadence reflects competitive market pressure: Google, Anthropic, and other labs are releasing models at similar frequency, and ChatGPT's default model is one of the primary user retention factors.
What It Means for Developers Using the API
If your application points to chat-latest in the API, it will automatically receive GPT-5.5 Instant. If you point to a specific model identifier like gpt-5.3-instant, you'll keep using that model until its retirement in three months.
OpenAI's recommendation for production is always to use specific model identifiers — not chat-latest — to avoid unexpected behavior changes when the default model updates. This kind of change is exactly why that practice matters.
The OpenAI models documentation keeps an updated list of available models and their planned retirement dates.