OpenAI has replaced GPT-5.3 Instant with GPT-5.5 Instant as the default model across all ChatGPT tiers, rolling out to Plus and Pro users immediately with Free, Business, and Enterprise access in coming weeks. The new model scores 81.2 on the AIME 2025 math benchmark (vs. 65.4 for the previous default) and 76 on MMMU-Pro multimodal reasoning (vs. 69.2), while specifically targeting reduced hallucinations in law, medicine, and finance — the three regulated verticals where AI citation authority is growing fastest.
A significant new feature: memory sources are now visible across all models, showing users exactly what past conversations, files, and Gmail context shaped a response. This transparency layer directly affects GEO — ChatGPT responses now carry an explicit citation chain that users can inspect and correct, making structured merchant content that gets correctly attributed far more durable than unstructured content that may be misattributed. API developers get access as "chat-latest"; GPT-5.3 remains via API for three months only.
TechCrunch · May 5, 2026