OpenAI has released GPT-5.2, its latest artificial intelligence model series, on December 11, 2025. The company describes it as the most capable model for professional knowledge work to date. The release includes three variants: GPT-5.2 Instant for faster responses, GPT-5.2 Thinking for complex tasks, and GPT-5.2 Pro for the most demanding questions.
According to OpenAI, the model shows improvements in creating spreadsheets, building presentations, writing code, understanding images, processing long documents, and handling multi-step projects. The company states that existing ChatGPT Enterprise users already report saving 40 to 60 minutes daily, with heavy users saving more than 10 hours weekly.
OpenAI tested GPT-5.2 Thinking on GDPval, an internal benchmark measuring knowledge work tasks across 44 occupations. The company claims the model beat or matched human professionals in 70.9 percent of comparisons, completing tasks at over 11 times the speed and less than 1 percent of the cost. Tasks included creating presentations, spreadsheets, and other professional artifacts. One benchmark judge commented that outputs appeared to have been done by a professional company with staff.
The model achieved 55.6 percent on SWE-Bench Pro, a software engineering evaluation testing real-world coding tasks across four programming languages. On SWE-bench Verified, which tests Python only, GPT-5.2 Thinking scored 80 percent. Early testers reported stronger performance in front-end development and complex user interface work.
OpenAI claims GPT-5.2 Thinking produces 30 percent fewer responses with errors compared to GPT-5.1 Thinking on a set of ChatGPT queries. The company emphasizes that users should still verify answers for critical tasks. The model also demonstrates improved long-context reasoning, achieving near 100 percent accuracy on certain tests involving 256,000 tokens.
For vision tasks, GPT-5.2 Thinking cuts error rates roughly in half on chart reasoning and software interface understanding, according to OpenAI. The model shows stronger spatial awareness when identifying components in images and their positions.
The model achieved 98.7 percent on Tau2-bench Telecom, a test measuring tool use across multi-turn customer support tasks. OpenAI states this enables more reliable end-to-end workflows for tasks like resolving support cases and coordinating actions across multiple systems.
In scientific domains, GPT-5.2 Pro scored 93.2 percent on GPQA Diamond, a graduate-level question benchmark covering physics, chemistry, and biology. GPT-5.2 Thinking solved 40.3 percent of problems on FrontierMath, an expert-level mathematics evaluation. The model also became the first to exceed 90 percent on ARC-AGI-1, a general reasoning benchmark, while reducing costs by approximately 390 times compared to the previous o3-preview model.
The release comes amid increased competitive pressure for OpenAI. CEO Sam Altman reportedly issued a code red alert internally last week, signaling a company-wide push to improve ChatGPT as Google’s Gemini app has grown to 650 million monthly active users compared to OpenAI’s 800 million weekly active users. Fidji Simo, OpenAI’s CEO of Applications, denied that the launch was accelerated, stating the company had worked on the release for many months. She confirmed that additional resources have been redirected to ChatGPT improvements.
OpenAI says it continued work on safe completions and strengthened responses to prompts indicating self-harm, mental health distress, or emotional reliance on the model. The company is rolling out an age-prediction system to automatically apply content protections for users under 18. Simo stated that adult mode, which will allow users over 18 to have different types of conversations, is planned for the first quarter of 2026.
ChatGPT users on paid plans will see GPT-5.2 rolling out gradually starting today. GPT-5.1 will remain available for three months under legacy models before being sunset. In the API, OpenAI has no current plans to deprecate GPT-5.1, GPT-5, or GPT-4.1 and will provide advance notice for any future deprecations.
Sources: OpenAI’s Announcement, OpenAI Docs, Wired, The Verge
