OpenAI has officially announced a major upgrade to its model series with the release of GPT-5.2, a model the company describes as the first to achieve performance that matches or surpasses human experts in complex cognitive tasks.

According to a technical report published by The Verge, the new "GPT-5.2 Thinking" model achieved remarkable results in the GDPval benchmark tests, outperforming human professionals by 70.9% in specialized tasks including advanced programming and strategic financial analysis.

System 2 Thinking Deep Reasoning Architecture: GPT-5.2 relies on a revolutionary reasoning architecture that allows it to pause and "think" before issuing a response, completely eliminating the problem of hallucinations in mathematical and logical calculations. The new system doesn't just predict the next word; it builds a tree of possibilities and solutions before selecting the most accurate and reliable path, making it ideal for use in scientific research and engineering.

A Major Advancement in Codex Processing via Codex 2: The report revealed that the specialized programming version, "GPT-5.2-Codex," achieved a new record in the SWE-Bench Pro benchmark, outperforming open-source models by 55%. The model has become capable of independently managing entire software projects, including writing tests, debugging, and restructuring code in multiple programming languages ​​simultaneously.

Turning Data into Actions via Built-in Agents

The most significant feature of this update is the form's ability to "use tools" independently (Agentic Tool-calling). Instead of simply typing responses, GPT-5.2 can now connect to external programs, such as spreadsheets and presentation software, to create and edit complete files on behalf of the user, transforming it from a digital assistant into an expert virtual employee.