OpenAI says GPT-5.5 Instant cuts hallucinated claims sharply versus GPT-5.3 Instant and improves factuality across high-stakes prompts. It also responds with a tighter, more natural tone while handling images, STEM questions, and web search more effectively.
For builders, the most relevant change is better personalization and clearer memory-source controls. That should make long-running assistant workflows easier to trust, especially when models rely on prior chats, files, or connected Gmail.
ChatGPT users will see the rollout automatically, and API users can access the model as chat-latest. If you rely on memory-heavy workflows, review the new memory sources view and update any prompts or automation assumptions.
Read Original Post →