OpenAI released GPT-5.5, its first fully retrained base model since GPT-4.5. It leads Terminal-Bench 2.0 at 82.7% and, more importantly for everyday users, produces 52.5% fewer hallucinations than its predecessor.
Why “fewer hallucinations” matters
Hallucinations — confident but wrong answers — are the #1 trust problem with AI tools. A 50%+ reduction means fewer made-up facts, citations and code that doesn’t exist. It doesn’t eliminate the issue, but it noticeably improves reliability for research and coding.
What’s new
- Full retrain: a fresh base model, not a fine-tune on top of GPT-5.
- Coding leadership: top Terminal-Bench 2.0 score among shipping models.
- More reliable answers: better calibration on what it does and doesn’t know.
What this means for you
- ChatGPT answers should require less fact-checking — but still verify anything important.
- For coding, GPT-5.5 is now a top-tier choice alongside Claude Opus 4.8 and Gemini 3.1 Ultra.
- If you canceled over hallucination frustration, this release is worth a fresh look.
FAQ
Is GPT-5.5 free? Expect tiered access — free users get limited use, Plus users get higher limits.
Should I still verify its answers? Yes. Fewer hallucinations is not zero hallucinations.