OpenAI rolled out GPT-5.5 Instant as the new default model for ChatGPT, promising a substantial drop in factual errors. Internal testing shows the model produces 52.5% fewer hallucinated claims and 37.3% fewer inaccurate answers on user‑flagged conversations. The company markets the upgrade as a step toward clearer, more trustworthy interactions.

Beyond accuracy, GPT-5.5 Instant trims excess. It asks fewer unnecessary follow‑up questions, pares down cluttered output, and scales back emoji use. Users report more concise replies that still retain the warmth that makes ChatGPT appealing.

The update also speeds up how the assistant pulls information from uploaded files, linked Gmail accounts and past chat histories. By stitching together these sources faster, the model reduces the need for users to repeat details, delivering a smoother, more personalized experience.

Memory sources give users control

Alongside the model change, OpenAI introduced “memory sources,” a transparency tool that shows which bits of stored context influence a response. Users can delete or edit entries they no longer want cited, and they can opt for temporary chats that never touch their memory store. The feature does not expose personal data to others, even when a chat is shared.

Memory sources will appear for all ChatGPT users on web and mobile “soon,” though the initial rollout focuses on Plus and Pro plans. Free, Go, Business and Enterprise tiers are slated to receive the model upgrade in the weeks ahead, with full access to the memory tool expected shortly thereafter.

OpenAI’s moves aim to address long‑standing criticism about AI hallucinations and opaque personalization. If the reductions in errors hold up in real‑world use, the company could set a new benchmark for conversational AI reliability.

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