OpenAI shifts ChatGPT's default model to GPT-5.5

OpenAI updated ChatGPT’s default model for all users to GPT-5.5 Instant on May 8.

The company said it gives shorter answers, uses prior context more effectively, and improves accuracy in areas such as medicine, law, and finance.

In internal tests, OpenAI said GPT-5.5 Instant produced 52.5% fewer hallucinated claims than GPT-5.3 Instant on high-stakes prompts.

The update also reduced inaccurate claims by 37.3% in conversations flagged by users for factual errors.

The company also said the model improved on image analysis, STEM questions, and deciding when to use web search, while asking fewer follow-up questions and reducing long or heavily formatted replies.

🔗 Source: OpenAI

🧠 Food for thought

Implications, context, and why it matters.

The new default model carries “High capability” preparedness risks

GPT-5.5 Instant is framed as an everyday tool, yet OpenAI classifies it as “High capability” in cybersecurity and Biological & Chemical Preparedness. Those categories cover cyberattack risks and the misuse of biological or chemical knowledge 1.
OpenAI says that status brings safeguards for those areas. The company bases the cybersecurity label on tests run at “xhigh” reasoning effort, an internal setting for deeper reasoning, even though the model ships at low reasoning effort 1.
OpenAI’s system card, a technical document on performance and risk, says the model is less robust against jailbreaks, attempts to make it ignore safety rules. OpenAI adds that the finding is directional rather than definitive 1.

OpenAI’s strategy is to make powerful AI an invisible utility

This release also sharpens OpenAI’s position against Google’s Gemini Flash and Anthropic’s Claude Haiku, models built for large workloads that need fast replies 2.
OpenAI has made a more accurate and conversational system the default for everyone, including free users. That change raises the baseline people can expect from ChatGPT 3.
The broader goal is to turn ChatGPT into a daily tool. The move also suggests OpenAI learned from the backlash after it pulled GPT-4o, which some users liked for its “personality” 4.