OpenAI has officially launched its newest large language model series, GPT-5.2, marking the second major update to the GPT-5 family since its launch in August. The rollout began on December 11, positioning the new models as the company’s “most capable model series yet for professional knowledge work.”
Accessing the New Model Series
GPT-5.2 is actually a series of models, each designed for a different balance of speed and reasoning:
- GPT-5.2 Instant: Fastest, best for everyday, lightweight tasks and rapid responses.
- GPT-5.2 Thinking: Best for deep work, complex reasoning, long-context tasks, and professional knowledge work.
- GPT-5.2 Pro: The most advanced model for the hardest scientific, logical, and coding challenges.
When Can You Use It?
The rollout is happening in phases, starting with paying customers.
- Paid Users First: Users on ChatGPT Plus, Pro, Business, Enterprise, and Edu plans are the first to gain access. These users can manually select the Instant or Thinking variants from the model picker at the top of the ChatGPT interface.
- GPT-5.2 Pro Access: The highest-tier model, GPT-5.2 Pro, is currently reserved only for Pro, Business, Enterprise, and Edu plans.
- Free Users: Eventually, all ChatGPT users should gain free access to GPT-5.2, though that access will be limited in message count per hour or day. The new models are also available now via the developer Application Programming Interface (API).
OpenAI is deploying the models gradually to ensure a smooth and reliable experience. If you are a paid user and do not see the option immediately, the company advises checking back later.
Major Performance Improvements
The new GPT-5.2 models demonstrate significant iterative improvements over their predecessors, GPT-5 and GPT-5.1. These advancements focus heavily on reasoning, coding, and mathematical precision, leading to significantly fewer hallucinations.
Key benchmark scores highlight the leap in capability:
- GDPval (Knowledge Work): GPT-5.2 Thinking scored 70.9%, a massive jump from GPT-5’s 38.8%. This reflects a model that performs at or above human expert levels on tasks like creating spreadsheets and presentations.
- SWE-Bench Pro (Coding): GPT-5.2 Thinking achieved a new state-of-the-art score of 55.6%, up from GPT-5’s 50.8%. This translates to a more reliable model for debugging production code and implementing feature requests.
- AIME 2025 Math: GPT-5.2 scored a perfect 100%, improving upon GPT-5’s 94%.
GPT-5.1 will remain available to paid users as a legacy model for three months before it is fully retired, cementing GPT-5.2 as the new default for advanced ChatGPT functionality.