On Monday, OpenAI officially released the new basic model GPT‑5.5 Instant and set it as the default model of ChatGPT to replace the previous GPT‑5.3 Instant. The company stated that while maintaining low-latency performance similar to that of the previous generation, GPT‑5.5 Instant has significantly converged on “illusion” issues in sensitive fields such as law, medical care, and finance, thus improving overall reliability.
OpenAI launched a full version of the GPT‑5.5 model last month. At that time, the official emphasized its improved capabilities in code generation and knowledge-based work tasks. Among the newly announced performance metrics, GPT‑5.5 Instant achieved a score of 81.2 on the AIME 2025 math test, a significant jump from the 65.4 score of the old model. On the multimodal inference benchmark MMMU‑Pro, the new model also leads the way, scoring 76.0 points, higher than the 69.2 points of the previous model.
This update particularly emphasizes the enhancement of context and memory management capabilities. GPT‑5.5 Instant can use its search tools to look back on past conversations, files, and Gmail content to provide users with more personalized answers. This feature is first available to Plus and Pro users on the web, and OpenAI plans to expand to mobile soon. The company also said that it will gradually roll out the feature to free version, Go Business and enterprise users in the coming weeks.
In terms of user-side experience, ChatGPT will uniformly display the "memory sources" on which answers rely in all models to help users understand where the answers come from. Users can delete obsolete memory sources and correct responses if they are found to be incorrect. OpenAI particularly emphasizes that when a user shares a conversation with others, the other party cannot see the details of the memory source to protect privacy.
For developers, GPT‑5.5 models will be available as “chat‑latest” via the API. Existing 5.3 versions will remain available as an option for paying customers for three months, allowing for a transition period for product migration and compatibility.
OpenAI has previously been controversial over its model offline strategy. When the company decided to stop offering the GPT‑4o model, it sparked a massive backlash from users. Many users highly identify with GPT‑4o’s “personality traits” and believe that it often affirms their choices, thereby building a sense of emotional dependence. On multiple petition platforms, undersigned users have described the model as their “best friend” or “mirror,” calling on OpenAI to retain access to GPT‑4o. Despite the ongoing controversy, GPT‑4o was officially deactivated in February 2026.