OpenAI on May 5 began rolling out GPT-5.5 Instant as the new default model for ChatGPT, replacing GPT-5.3 Instant for the consumer chatbot's approximately 900 million weekly users. The company is positioning the upgrade primarily around accuracy gains in domains where wrong answers carry real-world consequences — law, medicine, and finance — alongside notable jumps on standardized math and multimodal reasoning tests.
Hallucination cuts lead the pitch
OpenAI says GPT-5.5 Instant produces 52.5% fewer hallucinated claims than GPT-5.3 Instant on high-stakes prompts in sensitive areas, and reduces inaccurate claims by 37.3% on especially challenging conversations users had previously flagged for factual errors. The model is also more concise: OpenAI reports it uses roughly 30.2% fewer words and 29.2% fewer lines to communicate the same point, an efficiency gain that compounds across hundreds of millions of daily prompts.
On benchmarks, GPT-5.5 Instant scored 81.2 on the AIME 2025 math test compared to 65.4 for GPT-5.3 Instant, and 76 on the MMMU-Pro multimodal reasoning benchmark, up from 69.2. The company describes the model as preserving the low latency of its predecessor — a key constraint for the default consumer experience, where a perceptible slowdown in response time would degrade the product even if answers improved.
Memory and personalization shift
The rollout pairs the new model with an expanded personalization layer. ChatGPT can now use its search tool to refer back to past conversations, files, and Gmail to ground answers in user-specific context. To address obvious privacy and trust concerns, OpenAI is introducing a feature called "memory sources" that lets users see exactly which context the assistant pulled in for any given personalized response and adjust or remove that data.
The context management features are launching first for Plus and Pro users on the web, with mobile to follow. Free, Go, Business, and Enterprise users will gain access in the coming weeks.
Developer access and migration window
For developers, GPT-5.5 is exposed through the API under the "chat-latest" alias, which always points to the model OpenAI is currently using as the ChatGPT default. GPT-5.3 will remain selectable for paid API users for three months before being retired, giving teams a relatively short window to migrate evaluation suites, prompt libraries, and any latency-sensitive integrations.
Why it matters
GPT-5.5 Instant is not a flagship reasoning model — those announcements typically arrive separately — but as the default ChatGPT model it is the version most people will actually use. A double-digit reduction in hallucinations on legal, medical, and financial prompts, paired with measurably more concise outputs, is the kind of incremental change that has outsized impact on enterprise pilots, regulated-industry deployments, and the long tail of consumer use cases where accuracy quietly determines whether ChatGPT is trusted for the next task. The decision to fold tighter Gmail and file integration into the default experience also signals that OpenAI sees personalization, not raw capability, as the next axis of competition with Anthropic's Claude and Google's Gemini.
