OpenAI Launches GPT-5.2 to Automate Expert-Level Professional Work and AI Agents

OpenAI has officially launched GPT-5.2, a new series of frontier models engineered to transform professional knowledge work and autonomous agents. Released on December 11, 2025, the lineup introduces three distinct versions: Instant, Thinking, and a high-compute Pro tier. This update marks a strategic pivot toward economic utility, aiming to automate complex work products like financial models and software patches at expert levels.

The flagship Thinking model establishes a new industry standard for economic productivity. On the new GDPval benchmark, it outperformed or tied industry professionals in 70.9% of specified tasks across 44 different occupations. Users on ChatGPT Plus and Enterprise plans can now utilize these tools to generate sophisticated spreadsheets and presentations directly from simple prompts, saving an estimated 40 to 60 minutes of daily labor for average users.

Technical performance metrics show significant leaps in specialized fields like engineering and mathematics. GPT-5.2 Thinking achieved a record 55.6% on the rigorous SWE-Bench Pro, while the Pro variant reached a 54.2% score on the ARC-AGI-2 reasoning test. These models also feature a 30% reduction in hallucination rates compared to previous versions, making them more dependable for high-stakes research and analysis.

To support these advanced capabilities, OpenAI has introduced a tiered pricing structure in the API platform. The high-performance GPT-5.2 Pro model is priced at $21 per million input tokens and $168 per million output tokens, specifically targeting specialized scientific and mathematical research. Developers can also access the Thinking and Instant models at more competitive rates, with a 90% discount applied to cached inputs to improve long-term cost efficiency.

The development of GPT-5.2 was fueled by hardware collaborations with NVIDIA and Microsoft, utilizing advanced GPU clusters. This infrastructure supports an expanded context window, allowing the models to maintain coherence across massive documents and multi-file projects. Additional details on the rollout and safety measures can be found in the official OpenAI Product Release documentation. While the models show significant gains, the competition remains fierce in scientific evaluations. GPT-5.2 Pro achieved a 93.2% score on the GPQA Diamond science benchmark, narrowly trailing the 93.8% mark set by Google’s Gemini 3 earlier this year. This tit-for-tat dynamic highlights the aggressive pace of development among top AI labs as they vie for dominance in specialized reasoning.