OpenAI implements GPT-5.5 Instant as standard and reduces high-risk information distortions
OpenAI launched this Tuesday (May 5, 2026) GPT-5.5 Instant as the new ChatGPT standard model. It replaces GPT-5.3 Instant, which will remain available for another 3 months before being discontinued. The main change is the increase in response accuracy, with a reduction in errors and .
Internal testing shows that GPT-5.5 Instant generated 52.5% fewer responses with false information in high-risk situations, such as medicine, law and finance topics. In conversations already marked by users for containing errors, the new model had 37.3% fewer incorrect statements.
MORE CONCISE AND IMPROVED ANSWERS
According to OpenAi, the system also improved performance in everyday tasks. It now better answers science, technology, engineering and math questions, and analyzes images more accurately.
Another change is in the style of the answers. According to the company, GPT-5.5 Instant offers answers “leaner and more direct to the point” to avoid long texts, excessive formatting and unnecessary questions.
The update also improves the use of previous user information. The model can make better use of past conversations, files and connected accounts, such as Gmail, to generate more relevant responses. This reduces the need to repeat data.
Among the new features is the “memory sources” feature, which shows what information was used to personalize each response. The user can review, correct or delete this data. The tool does not share this information with other people and only displays the most relevant elements.
The most advanced customization features are starting to be released to users of the Plus and Pro plans in the web version. Expansion to mobile apps and other plans is expected in the coming weeks. “Memory sources” is now being made available to all users on the web, with variations depending on the region. The company did not specify which ones.
