OpenAI Introduces GPT-5.5 Instant, Replacing GPT-5.3 Instant

(AsiaGameHub) –   On Tuesday, OpenAI launched GPT-5.5 Instant as the new default ChatGPT model, taking the place of GPT-5.3 Instant while retaining its core focus on low latency.

---

Key Details to Note

• GPT-5.5 Instant earned a score of 81.2 on the 2025 AIME exam, a notable increase from the 65.4 mark recorded by GPT-5.3 Instant.
• OpenAI states that the new model cuts down on hallucination occurrences across the law, medicine and finance fields.
• Web-based Plus and Pro subscribers will be the first to receive access to expanded context features.

---

GPT-5.5 Instant Delivers Quicker Responses With Enhanced Context Capabilities

OpenAI is updating ChatGPT’s default model once again, but the company hopes users will perceive the upgrade as practical rather than disruptive. GPT-5.5 Instant retains the fast response style of GPT-5.3 Instant, while bringing improvements to reasoning, coding and knowledge work performance.

The benchmark performance gains are unambiguous. GPT-5.5 Instant hit a score of 81.2 on the 2025 AIME math test, compared to the 65.4 achieved by the older Instant model. It also notched a 76 score on MMMU-Pro, outperforming the 69.2 result posted by GPT-5.3 Instant.

For daily users, context handling improvements may have a greater impact than benchmark score increases. GPT-5.5 Instant can leverage search tools to reference past conversations, files and Gmail content, delivering more personalized responses when access permissions are enabled. Web-based Plus and Pro users will get this feature first, with mobile access rolling out at a later date. OpenAI also plans to expand access to Free, Go, Business and enterprise tier users in the coming weeks.

Moving forward, OpenAI will display memory sources across all ChatGPT models. Users can check the origin of any given answer, delete outdated memory sources, or correct inaccurate entries. Shared conversations will not expose these memory sources to other users.

Developers will be able to access GPT-5.5 via the API under the chat-latest designation. Paid users will still have the option to select GPT-5.3 for a three-month window before that option is phased out entirely.

Model adjustments have sparked user frustration on prior occasions. OpenAI faced widespread backlash after retiring GPT-4o, a model many users preferred for its warmer, more approachable tone. Some signatories of a petition opposing the retirement described the model as their “best friend” or “a mirror”. OpenAI still proceeded to deprecate GPT-4o in February 2026 regardless of the pushback.