OpenAI has released GPT-5.2, a new family of frontier models available across ChatGPT and the API, designed to enhance professional work, long-running agents, and complex knowledge tasks.
The release includes three variants: Instant for everyday assistance, Thinking for multi-step work and agents, and Pro for hard technical and analytical tasks. GPT-5.2 Thinking achieves 80.0 percent on SWE-bench Verified and near 100 percent accuracy on the OpenAI MRCRv2 long-context benchmark. The model also demonstrates improved vision quality and scientific reasoning, with GPT-5.2 Pro scoring 93.2 percent on GPQA Diamond.
This update positions GPT-5.2 Thinking as the new default workhorse for coding and knowledge work, offering significant accuracy improvements over GPT-5.1 while maintaining comparable token limits.