OpenAI replaced GPT-5.3 Instant with GPT-5.5 Instant as the default ChatGPT model on 5 May 2026. The company says the upgrade cuts hallucination rates in legal, medical and financial queries by roughly half while keeping per-token latency at sub-300 milliseconds.
OpenAI, Google, Microsoft, Amazon, Oracle, Nvidia, SpaceX and Reflection AI are in. Anthropic is out, after refusing to let Claude be used for autonomous weapons and mass surveillance.
Released on 24 April 2026, GPT-5.5 is positioned for agentic coding, knowledge work and scientific research. It costs more per token than GPT-5.4 but uses fewer tokens to reach higher-quality outputs.
District Judge Sidney H. Stein upheld the magistrate order requiring OpenAI to produce 20 million anonymised ChatGPT logs — 0.5% of its preserved data — in the consolidated copyright suits.
Anthropic's latest model expands its constitutional AI framework to over 200 principles. OpenAI's parallel "meta-feedback" technique — humans critiquing model reasoning, not just outputs — is reportedly cutting reward hacking.