OpenAI has just officially released GPT-5.5 Instant, the new platform model chosen as default for ChatGPT, replacing the previous GPT-5.3 Instant version.
The upgrade focuses on reducing the phenomenon of "ghosts", which is a common problem of AI when providing misleading information, especially in sensitive areas such as law, healthcare and finance.
According to OpenAI, GPT-5.5 Instant still maintains fast response speed, low latency similar to the old version, but significantly improves accuracy and reliability.
This is considered an important step to balance performance and user experience in real-world applications.
A month earlier, OpenAI announced the GPT-5 model with many upgrades in programming capabilities and processing complex intellectual tasks.
This Instant version inherits those improvements, while optimizing for the daily usage environment.
The evaluation results show that GPT-5.5 Instant achieved 81.2 points in the AIME 2025 math test, a sharp increase compared to 65.4 of the predecessor version.
In the MMMU-Pro multi-modal reasoning evaluation standard, the model also achieved 76 points, exceeding the previous 69.2. These figures reflect a clear improvement in information processing and reasoning capabilities.
A notable highlight is the ability to manage context. GPT-5. 5 Instant can use the search engine to retrieve previous conversations, Gmail files or emails to give more personalized answers.
This feature is initially for Plus and Pro users on the web platform, and is expected to soon expand to mobile devices as well as other user groups.
In addition, ChatGPT is also supplemented with the ability to display the memory source, allowing users to know which data the answer is formed from.
Users can edit or delete outdated information sources to ensure accuracy. Notably, when sharing a conversation, the recipient will not be able to view these memory sources, helping to protect privacy.
For the developer community, GPT-5.5 will be provided via an API called "chat-latest". Meanwhile, the GPT-5.3 version will only be maintained for paid users for three months before being completely removed.
The replacement of the default model once again shows OpenAI's fast updating strategy, although it once caused controversy. However, OpenAI remains steadfast in its innovation path, towards more accurate and secure AI systems.
