OpenAI has released GPT-5, its first "unified" AI model that combines the reasoning capabilities of its o-series with the speed of the GPT series. Released on Thursday, GPT-5 becomes the default model for all ChatGPT users, including those on free tiers, marking a shift toward agentic AI systems.
- GPT-5 features a real-time router that dynamically decides whether to respond quickly or engage in extended reasoning.
- On SWE-bench Verified, GPT-5 scores 74.9%, slightly edging out Anthropic's Claude Opus 4.1 (74.5%).
- The model reduces hallucinations significantly, with a rate of 4.8% compared to 22% for o3.
- Developer APIs will offer three sizes: gpt-5, gpt-5-mini, and gpt-5-nano, along with verbosity controls.
This release makes advanced AI reasoning available to free users for the first time while positioning ChatGPT as a tool capable of completing complex tasks on behalf of users.