GPT-5.2 released; A quick start guide to using the most powerful OpenAI model

OpenAI has announced that the new version of GPT-5.2 (GPT-5.2) is already activated and access to it will be gradually activated for users from December 11. This version is the second major update of the GPT-5 (GPT-5) chat family; The model that was introduced after a long wait on August 7.

In a blog post announcing the new GPT chat series, OpenAI described GPT 5.2 as “the most powerful set of AI models to date for professional knowledge-based tasks.”

GPT 5.2 is actually a collection of models and comes as part of the larger Chat GPT 5 family. The models in this series are:

• GPT-5.2 Instant
• GPT-5.2 Thinking
• GPT-5.2 Pro

When can GPT 5.2 did you use

Officially released on December 11, GPT 5.2 is now live, but may not yet appear in the list of GPT chat models for all users. According to the usual policy of Open AI, users with subscriptions including Plus, Pro, Go, Business and Enterprise will get access to this model first.

Even for users with a subscription, activation is done gradually. So if this model is still not visible in the API or JPT chat, just check back later.

OpenAI has written in the explanation of this process:

We’re rolling out GPT 5.2 gradually to keep GPT Chat smooth and reliable. If you don’t see it at first, please try again later. In GPT Chat, the GPT 5.1 model will remain available to paid users under the legacy title for three months, after which it will be retired.

Users with a paid subscription can check if this model is enabled for them by visiting ChatGPT.com. Finally, all users will have access to GPT 5.2, although this access will be limited in the free plan.

Benchmarks and performance GPT 5.2

OpenAI has released a system map for the new model, which shows that GPT 5.2 has improved in many key metrics over GPT 5 and GPT 5.1. The company also says that the hallucination rate of the model has been reduced in version 5.2.

Some of the announced results are:

• GDPval:
  • GPT-5.2: 70.9%
  • GPT-5: 38.8%
• SWE-Bench Pro:
  • GPT-5.2: 55.6%
  • GPT-5: 50.8%
• AIME 2025 (Mathematics):
  • GPT-5.2: 100%
  • GPT-5: 94%
• FrontierMath (Level 1 to 3):
  • GPT-5.2: 40.3%
  • GPT-5: 31.0%