OpenAI has introduced its GPT-5.2 model lineup. As of December 12, the Instant, Thinking, and Pro versions are available to all users.
GPT-5.2 is now rolling out to everyone.https://t.co/nfubPwnIIw
— OpenAI (@OpenAI) December 11, 2025
“Currently, ChatGPT Enterprise users save an average of 40-60 minutes daily thanks to AI, with active users saving more than 10 hours weekly. We developed GPT-5.2 to unlock even greater economic potential for people,” the startup’s blog states.
Thinking has shown high results in several tests.
The model has reached expert level in the GDPval benchmark, which evaluates the performance of intellectual tasks across 44 professions. Test tasks include creating presentations and working with spreadsheets.
The “Thinking” version of the LLM completes GDPval benchmark tasks 11 times faster than human experts, with the cost of task completion being less than 1% of a specialist’s expenses.
The company emphasized that GPT-5.2 Thinking “raises the bar for professional work.” It demonstrates:
- modern reasoning methods considering long context;
- significant improvements in spreadsheet generation and analysis;
- success in creating slide shows.
The GPT-5.2 Instant version is designed for everyday work and learning. It features a warm and conversational style, clear explanations highlighting key information, improved step-by-step guides, and high-quality translation of technical information.
GPT-5.2 Pro is positioned as the most powerful solution for complex queries. The neural network shows high performance in specialized fields, including programming and scientific research.
“GPT-5.2 is part of the ongoing process of model improvement. We continue to work on known issues such as unwarranted refusals and delays to make the product more useful,” OpenAI emphasized.
GPT-5.1 will remain available to paying users for three months.
Programming
GPT‑5.2 Thinking set a new record in the SWE-Bench Pro test — 55.6%. It analyzes the model’s ability to work with four languages.
The SWE-bench Verified test also achieved a high value — 80%.
“For everyday professional use, this means the model more reliably debugs code, implements requests for new features, refactors large codebases, and fixes from start to finish with less manual intervention,” states the OpenAI blog.
GPT-5.2 Thinking performs better in frontend development and creating complex and atypical interfaces compared to GPT-5.1 Thinking.
“GPT-5.2 is the biggest leap for GPT models in agent programming since GPT-5 and is the best solution in its price range,” noted OpenAI.
Hallucinations and Context
GPT‑5.2 Thinking hallucinates less compared to GPT‑5.1 Thinking. The model is more reliable in everyday information work, conducting research, writing text, analyzing, and supporting decision-making.
GPT‑5.2 Thinking “sets a new standard” in long-context reasoning. It achieved leading scores in OpenAI MRCRv2 — a test that checks the model’s ability to integrate information distributed across long documents.
In real-world deep analysis tasks requiring related information across hundreds of thousands of tokens, GPT‑5.2 Thinking is “significantly more accurate” than GPT‑5.1 Thinking.
Vision
GPT‑5.2 Thinking is OpenAI’s most powerful visual perception model. It reduces errors in diagram analysis and understanding software interfaces by about half.
The neural network can more accurately interpret dashboards, screenshots, technical graphs, and reports.
The example shows that GPT-5.2 successfully identifies key areas and delineates object boundaries. In contrast, GPT-5.1 highlights only individual fragments, showing weak understanding of spatial structure.
Although both models make errors, version 5.2 handles image analysis noticeably better.
In December, OpenAI planned the release of GPT-5.2 as a response to the growing popularity of Google’s Gemini.