OpenAI has officially launched GPT-5.2, describing it as its most capable model series yet for professional knowledge work and complex problem-solving. The update arrives just weeks after the release of GPT-5.1 and is aimed at raising the bar for coding, science, and agent-style workflows.
According to OpenAI, GPT-5.2 sets a new standard on the GDPval benchmark, outperforming industry professionals on well-specified knowledge work tasks across 44 occupations. In software development, the "Thinking" variant of the model scored 80% on SWE-bench Verified and 55.6% on the more demanding SWE-bench Pro, allowing it to tackle multi-step coding projects, debug production code, and implement feature requests with less human oversight.
The release is especially relevant for Apple users. OpenAI's models power advanced queries for Siri and other Apple Intelligence features, and GPT-5.2 is meant to improve the handling of complex requests made from Apple devices. It also builds on OpenAI's growing presence on the Mac, following the launch of ChatGPT Atlas and the acquisition of the Mac app Sky for deeper system integration.
Reliability is a key focus this time. OpenAI says GPT-5.2 produces fewer incorrect answers than GPT-5.1, with a 30% relative reduction in factual errors on internal tests. Visual understanding has also been improved, roughly halving error rates on tasks involving charts, software interfaces, and other structured imagery. That should make the model more useful for interpreting dashboards, technical diagrams, product screenshots, and other visual data.
GPT-5.2 is available in three main versions: GPT-5.2 Instant for quick, everyday tasks; GPT-5.2 Thinking for deeper reasoning and coding; and GPT-5.2 Pro for the most demanding scientific and technical questions. The new models are rolling out starting today for paid ChatGPT Plus, Pro, and Enterprise subscribers, while GPT-5.1 will remain available as a legacy option for three months.