Free ChatGPT: 52% Fewer Hallucinations Now

52.5% fewer hallucinated statements. That's the headline improvement in GPT-5.5 Instant — the default model powering free ChatGPT since May 2026. A separate update also cut factual errors by 37.3%. OpenAI rolled both changes out to all users without requiring any subscription upgrade.

What actually changed

The update focused on something deceptively hard to get right: understanding the intent behind a request, not just the literal words. GPT-5.5 Instant now tracks context more reliably across multi-turn conversations — if you're in a long back-and-forth, the model holds the thread instead of drifting. When you push back or clarify and the first answer missed the mark, the new version recalibrates rather than stubbornly restating itself.

Contextual relevance improved in practical ways too. Ask for restaurant recommendations and the model now works in your location without you explicitly spelling it out. Responses also got less templated — more naturally formatted prose where that fits, structured lists where that's useful, rather than forcing every answer into the same bulleted shape.

Why this matters

Free ChatGPT reaches hundreds of millions of people who use it as a first-stop research tool, writing aid, and quick-answer machine. The accuracy gap between free and paid AI has been real — and frustrating for anyone burned by a confident, wrong answer. A 52% drop in hallucinated statements is a meaningful shift on that trust spectrum. Not a solved problem, but a genuinely different place on the curve.

The broader strategic signal is just as significant: OpenAI is investing in raising the quality floor, not just the ceiling. The frontier models get the headlines, but the free tier — which is most people's actual daily relationship with AI — quietly became materially more reliable. At the scale of hundreds of millions of users, that compounds.