OpenAI rolled out a new default model for its ChatGPT service on Tuesday, branding it GPT-5.5 Instant. In internal evaluations, the model produced 52.5% fewer hallucinated claims than the earlier GPT-5.3 Instant when faced with high‑stakes prompts in fields such as medicine, law and finance. Users who flagged factual errors also saw a 37.3% drop in inaccurate statements, the company reported.

Fewer hallucinations, sharper answers

The reduction in fabricated content marks a significant step for a technology that has long struggled with factual reliability. OpenAI said the new model delivers “significant improvements in factuality across the board,” delivering tighter, more to‑the‑point responses while avoiding gratuitous emojis. Beyond factual accuracy, GPT-5.5 Instant expands everyday capabilities, including better analysis of image uploads and more discerning use of web searches when a question calls for up‑to‑date information.

Personalization and memory sources

OpenAI also introduced a “memory sources” feature that lets the chatbot reveal which pieces of context influenced a personalized reply. Users can delete or correct that context if needed, giving them more control over the data that shapes their interactions. The feature launches now for all ChatGPT consumer plans on the web, with mobile rollout slated for the near future.

Personalization gains are rolling out first to Plus and Pro subscribers on desktop, while free, Go, Business and Enterprise tiers will receive the enhancements “soon,” according to the company’s rollout plan. The new model also promises better integration with prior chats and even Gmail content, allowing the assistant to tailor responses based on a broader history of user interaction.

OpenAI will keep GPT-5.3 Instant available for three months as a fallback option before retiring it. This mirrors past transitions, where the company gave users time to adjust before older versions were removed. The gradual phase‑out aims to smooth the shift for developers and power users who rely on the previous model’s behavior.

Industry observers note that the emphasis on reduced hallucinations could raise the bar for competing AI products. Google’s Gemini, for instance, is reportedly investing heavily in similar factuality improvements. OpenAI’s public claim of a 52.5% reduction, however, rests on internal testing, and independent verification will likely follow as the model reaches a broader audience.

For now, OpenAI’s message is clear: the next generation of ChatGPT will be more reliable, more transparent about its reasoning, and more attuned to individual user needs. Whether the promised gains hold up under real‑world scrutiny remains to be seen, but the rollout marks the most substantial upgrade to the service’s core engine to date.

Used: News Factory APP - news discovery and automation - ChatGPT for Business