GPT 5.2: What’s new
OpenAI has introduced GPT 5.2 as its most powerful model for real professional tasks. The system is designed to support people who spend much of their day preparing documents, building models, writing code, or managing information inside large projects, Qazinform News Agency correspondent reports, citing OpenAI.
GPT 5.2 is designed to deliver finished work products rather than short answers. It generates full spreadsheets, polished presentations, code patches, and detailed project plans with more structure and fewer mistakes than earlier versions. OpenAI reports that the model works faster than human specialists on structured tasks and produces outputs that judges often rate as expert level.
Stronger performance
The new model sets high scores on evaluations that simulate real office work. On the GDPval benchmark, which measures tasks across more than forty occupations, GPT 5.2 beats or matches industry specialists in most cases. The system also shows clear gains in investment banking-style spreadsheet modeling, producing better-formatted and more consistent financial sheets.
A leap in coding
For software engineering, GPT 5.2 performs strongly on tests that require it to read real codebases and produce working patches. It shows more reliable debugging, feature implementation, and refactoring across several programming languages. Early users note visible improvement in front-end and interface-related tasks, including complex layouts and three-dimensional elements, which it can now produce from a single prompt.
Better reasoning with long documents
OpenAI says GPT 5.2 is its most capable model for long-context work. It can keep track of information spread across very large documents, which supports tasks such as reviewing lengthy reports, analyzing transcripts, integrating multiple sources, and managing large project files.
Improved vision and image understanding
The update brings a more accurate grasp of charts, dashboards, diagrams, and software interfaces. The model identifies elements on a page with greater spatial awareness and performs better on tasks that require chart reading or interpreting visual layouts.
Stronger tool use for full workflows
GPT 5.2 coordinates multi-step workflows with higher reliability. Tests show better performance on tasks that require the model to call external tools, switch between systems, or handle long customer support cases from beginning to end. The model can manage sequences such as rebooking travel, gathering data, preparing a summary, and delivering a final reply without dropping steps.
Gains in science and mathematics
The model shows higher accuracy on graduate level questions in physics, chemistry, and biology, along with stronger results in advanced mathematics benchmarks.
Availability
GPT 5.2 is rolling out to ChatGPT Plus, Pro, Business, and Enterprise users, with Instant, Thinking, and Pro versions. It is also available in the API. OpenAI says it will continue supporting earlier GPT 5 models for several months in ChatGPT and for longer in the API.
Earlier, Qazinform News Agency reported that a coalition of US state attorneys general had warned the largest AI companies that they must address unsafe chatbot behavior.