Overview
- OpenAI set GPT-5.5 Instant as ChatGPT’s default on Tuesday, exposed it to developers as chat-latest, and will keep GPT-5.3 Instant available to paid users for three months.
- The company says the new model makes 52.5% fewer hallucinated claims on high-stakes prompts and 37.3% fewer inaccurate statements on user-flagged chats than GPT-5.3 Instant, based on internal tests.
- Personalization can now draw on past chats, uploaded files, and connected Gmail, and a memory sources view shows what data shaped a reply and lets users delete or correct it.
- OpenAI classifies GPT-5.5 Instant as High Capability in cybersecurity and biological domains and says it added extra safeguards at deployment.
- OpenAI cites gains on benchmarks like AIME math and MMMU‑Pro, and early hands-on reviews report tighter answers with fewer emojis and fewer unnecessary follow-ups.