Reduced Hallucinations
OpenAI has rolled out a noteworthy advancement in its artificial intelligence capabilities with the introduction of GPT-5.5 Instant. This new foundational
model is now the standard for the ChatGPT service, and the company proudly reports that it has achieved a remarkable reduction in generated inaccuracies. Specifically, when tested with prompts requiring high levels of precision in domains such as medicine, law, and finance, GPT-5.5 Instant produced 52.5% fewer instances of 'hallucinated' or fabricated claims compared to its predecessor, GPT-5.3 Instant. This substantial improvement underscores OpenAI's dedication to making its AI tools more dependable and accurate for users engaging with complex information.
Benchmark Performance Gains
Beyond its enhanced factual accuracy, the new GPT-5.5 Instant model demonstrates superior performance across a range of challenging benchmarks. In the AIME 2025 mathematics test, the model achieved an impressive score of 81.2, a significant leap from the previous version's score of 65.4. Furthermore, it excelled in the MMMU-Pro multimodal reasoning benchmark, scoring 76.0, which surpasses the older model's score of 69.2. These elevated scores signify a robust enhancement in the model's analytical and comprehension abilities, reflecting OpenAI's ongoing efforts to refine its AI technology and deliver progressively better outcomes for diverse applications.
Enhanced Context Management
A key focus in the development of GPT-5.5 Instant has been the improvement of its context management capabilities. The model is now equipped to utilize its search functionality to access and reference previous conversations, uploaded files, and even Gmail data, enabling it to generate more personalized and contextually relevant responses. This advanced feature will initially be available to Plus and Pro subscribers on the web, with plans for a mobile app rollout in the near future. OpenAI intends to extend this enhanced contextual understanding to Free, Go Business, and enterprise users within the coming weeks, broadening its accessibility.
Transparency Features
In addition to the core model upgrades, ChatGPT is implementing other significant changes to enhance user experience and transparency. Across all models, ChatGPT will now display the sources from which its answers are derived. This new feature aims to empower users by allowing them to understand the origin of the AI's information. Moreover, users will have the ability to manage these sources, with options to delete or correct any outdated or inaccurate references. The underlying GPT-5.5 model is accessible via API under the name 'chat-latest,' while the previous GPT-5.3 model will remain available for paid users for an additional three months.
