OpenAI released its new generation foundational model GPT-5.5Instant on Tuesday and announced that it will replace GPT-5.3Instant as the default model in ChatGPT. While maintaining low latency, the model focuses on reducing cognitive biases in sensitive fields such as law, medicine, and finance. This update continues the company's overall direction introduced last month with the GPT-5.5 series model, which emphasizes enhanced coding capabilities and performance in knowledge-based tasks.

openai

According to benchmark tests, GPT-5.5Instant shows significant performance improvements: it scored 81.2 on the AIME2025 math test, significantly higher than the previous generation's 65.4; and it scored 76 on the MMMU-Pro multimodal reasoning test, a clear improvement over the previous generation's 69.2. This demonstrates an enhanced ability in complex reasoning and cross-modal understanding.

In terms of product features, this upgrade positions "context management" as one of the core capabilities. GPT-5.5Instant can retrieve historical conversations, documents, and Gmail emails through search tools to generate more personalized responses. This feature is now available to web version Plus and Pro users, and will be expanded to mobile devices and the Free, Go Business, and enterprise user systems in the coming weeks.

At the same time, ChatGPT introduced a "visualized memory source" mechanism, allowing users to view the historical information sources cited by the model's answers and support deleting or correcting related memories. In terms of privacy, OpenAI clearly stated that when sharing chat records, memory sources will not be exposed, reducing concerns about data usage.

On the developer side, GPT-5.5 will be provided via API in the form of "chat-latest," while GPT-5.3Instant will remain as a paid option for a three-month transition period.

Notably, OpenAI's past model replacement strategies have sparked user controversy. Previously, when GPT-4o was phased out in February 2026, it faced petitions due to its "high empathy style" creating user emotional dependence. This background makes the replacement path of GPT-5.5Instant face higher demands in terms of product experience continuity and user psychological expectations.

Overall, the release of GPT-5.5Instant marks a further balance between performance, controllability, and personalization capabilities of foundational models, and also reflects the platform's accelerated efforts to promote model standardization and unified access strategy.