OpenAI has rolled out GPT-5.5 Instant as the new default model for ChatGPT, replacing GPT-5.3 Instant across its platform. The company says the latest model significantly cuts down on hallucinations in critical areas including law, medicine, and finance while keeping the quick response times users expect.
This release comes as AI companies face mounting pressure to create more reliable and accurate language models. With ChatGPT now handling millions of queries daily across professional settings, reducing false information has become a top priority for maintaining user trust and expanding into enterprise markets.
According to TechCrunch, the new model shows impressive performance gains across key benchmarks. GPT-5.5 Instant scored 81.2 on the AIME 2025 math test, a substantial jump from the previous model’s 65.4 score. The model also outperformed its predecessor on multimodal reasoning tasks, achieving 76 points versus 69.2 on the MMMU-Pro benchmark.
The biggest change comes in how the model handles context and memory. GPT-5.5 Instant can now search through past conversations, uploaded files, and even Gmail to provide more personalized responses. This represents a major shift toward AI assistants that remember and learn from user interactions over time.
Key features of the new model include:
- Enhanced context management that references previous conversations and files
- Improved performance in coding and knowledge work tasks
- Visible memory sources so users can see where answers come from
- Ability to delete or correct outdated information sources
- Privacy protection for memory sources in shared chats
The memory and personalization features will initially be available to Plus and Pro subscribers on the web platform. OpenAI plans to bring these capabilities to mobile apps soon, followed by Free tier users and business customers in the coming weeks.
For developers, the company is making GPT-5.5 available through its API under the name “chat-latest.” Paid API users can still access the older GPT-5.3 model for three months during the transition period.
This rollout comes after OpenAI faced significant backlash earlier this year when it retired GPT-4o in February 2026. Users had formed emotional connections to that model’s personality and communication style, with some describing it as a “best friend” and launching petitions to save it. The controversy highlighted how users can develop preferences for specific AI model behaviors and responses.
The introduction of visible memory sources appears designed to give users more control and transparency over their AI interactions. Users can now see exactly what information ChatGPT used to generate responses and make corrections when the AI gets something wrong. However, these memory sources remain private – they won’t be visible to others when chat conversations are shared.
This update reflects broader industry trends toward more reliable and contextually aware AI systems. As language models become integrated into professional workflows and personal productivity tools, companies like OpenAI are focusing on reducing errors while making AI assistants more helpful through better memory and personalization features.
