GPT-5.5 Instant: Benchmarking the 52% Hallucination Reduction (the-decoder.com)

OpenAI has launched GPT-5.5 Instant as the new default model for ChatGPT, significantly reducing hallucinations by 52.5% on high-risk topics such as medicine, law, and finance. The model also shows impressive benchmark improvements, with accuracy soaring on tests like AIME 2025—from 65.4% to 81.2%—and GPQA, which measures PhD-level scientific reasoning, jumping from 78.5% to 85.6%. The update not only aims to provide more reliable information but also tightens responses by decreasing unnecessary verbosity, leading to shorter and more effective answers. A notable feature in this update is "memory sources," which allows users to see the personal context—such as past conversations or uploaded files—that informed a response. This feature enhances the personalization of interactions, although it currently requires a Plus or Pro subscription for full access, with broader rollout planned for other user tiers. Overall, GPT-5.5 Instant represents a significant step forward in making AI interactions more accurate and user-friendly, catering to the needs of both everyday users and professionals across various disciplines.