GPT-5.5 Instant: default model update with major factuality and personalization gains
Key Points
- GPT-5.5 Instant is now default
- 52.5% fewer hallucinations on high‑stakes prompts
- More personalized, shorter responses (~30% fewer words)
Summary
On 2026-05-05 OpenAI updated ChatGPT’s default model to GPT-5.5 Instant. The release focuses on improved factuality, tighter and clearer answers, better multimodal and STEM capabilities, and more effective use of past-chat and connected data for personalization. Responses are generally shorter and ask fewer unnecessary follow-ups.
Key Points
- Default model: GPT-5.5 Instant replaces previous Instant model for all users.
- Factuality gains: 52.5% fewer hallucinated claims on high-stakes prompts (medicine, law, finance) vs GPT-5.3 Instant; 37.3% reduction in inaccurate claims on flagged challenging conversations.
- Conciseness: ~30.2% fewer words and ~29.2% fewer lines on example advice prompts while keeping tone and utility.
- Capability improvements: better handling of photo/image uploads, improved STEM reasoning, and smarter decisions about when to use web search.
- Personalization: more effective use of past chats, files, and connected Gmail when permitted, with faster context search and conservative decisions about when to apply personalization.
- Behavior: fewer unnecessary follow-ups, less overformatting, more natural conversational tone.
Action items for engineers
- Run regression tests for high-stakes domains (medicine, law, finance) and verify factuality metrics against existing baselines.
- Re-evaluate prompt templates and system messages for brevity and clarity; consider tightening verbosity constraints.
- Test multimodal workflows (image/photo inputs) and STEM/math chains for correctness and edge cases.
- Validate personalization controls and privacy gating when past-chat/files/Gmail are used; ensure audits/logging are intact.
- Monitor production telemetry for hallucination rates, user follow-up frequency, and any latency/regression issues after rollout.
Notes
This release is positioned as an incremental but broad improvement— prioritize validation in your critical user flows and high-stakes evaluation suites.