OpenAI launches GPT‑5.5 Instant as new default model for ChatGPT
Key Points
- OpenAI released GPT‑5.5 Instant on Tuesday, making it the default model for ChatGPT.
- The new model reduces hallucinations in law, medicine and finance while keeping low latency.
- GPT‑5.5 scores 81.2 on the AIME 2025 math test, up from 65.4 for GPT‑5.3.
- Multimodal reasoning benchmark MMMU‑Pro improves from 69.2 to 76.
- Context‑management lets the model reference past chats, files and Gmail for personalized answers.
- Source attribution appears with each response; users can delete or correct references.
- Available now to Plus and Pro users on the web; broader rollout to other tiers planned.
- Developers can access GPT‑5.5 via the API as “chat‑latest,” with GPT‑5.3 remaining for three months.
OpenAI announced on Tuesday that its latest foundation model, GPT‑5.5 Instant, will replace GPT‑5.3 Instant as the default engine behind ChatGPT. The upgrade targets reduced hallucinations in high‑risk domains such as law, medicine and finance while preserving the low‑latency performance users expect. GPT‑5.5 also scores higher on math and multimodal reasoning benchmarks, and introduces a context‑management feature that lets the model reference past chats, files and Gmail. The changes roll out to Plus and Pro users on the web now, with broader availability slated for mobile and other plans in the coming weeks.
OpenAI unveiled GPT‑5.5 Instant on Tuesday, positioning the new model as the default behind ChatGPT and signaling a shift toward more reliable, domain‑aware AI assistance. The company highlighted the model’s ability to curb hallucinations in sensitive areas—including legal advice, medical information and financial guidance—without sacrificing the rapid response times that have become a hallmark of the chat service.
GPT‑5.5 builds on the foundation laid by its predecessor, GPT‑5.3 Instant, but delivers measurable gains across several benchmark tests. In the AIME 2025 math competition, the model posted an 81.2 score, a stark improvement over the 65.4 achieved by GPT‑5.3. Multimodal reasoning also saw a jump, with the MMMU‑Pro benchmark rising from 69.2 to 76. These figures suggest that the model not only answers questions more accurately but does so with a better grasp of visual and contextual cues.
One of the most visible enhancements is the new context‑management capability. GPT‑5.5 can now tap into a user’s prior conversations, uploaded documents and even Gmail messages to generate answers that feel personalized and grounded in real‑world data. The feature is currently live for Plus and Pro subscribers on the web, and OpenAI plans to extend it to mobile platforms shortly. A broader rollout to Free, Go Business and enterprise tiers is slated for the coming weeks.
Transparency is another focus. ChatGPT will display the sources it consulted for each response, allowing users to verify the origin of information. Users can delete outdated references or correct them if the answer proves inaccurate. Importantly, when a chat is shared, the recipient will not see the underlying memory sources, preserving privacy.
Developers will also see changes. The GPT‑5.5 model is accessible via the API under the name “chat‑latest,” while GPT‑5.3 remains an option for the next three months for paid users who need a transition period. This dual‑offering approach aims to smooth the shift for applications that depend on the older model’s behavior.
The release follows a controversial episode earlier this year when OpenAI retired its GPT‑4o model despite vocal user backlash. Some users described GPT‑4o as a “best friend” because of its conversational style. OpenAI’s decision to deprecate that model in February 2026 sparked petitions and criticism, underscoring the emotional connection many users have with AI personalities. By contrast, the company frames GPT‑5.5’s launch as a technical upgrade rather than a personality shift.
OpenAI’s rollout strategy reflects a balance between innovation and user trust. While the new model promises higher accuracy and richer context handling, the company continues to offer legacy options and transparent source attribution. As the AI landscape evolves, the firm appears intent on keeping performance gains aligned with user expectations and regulatory scrutiny.