OpenAI Released GPT-5.2: A New Model for Professional Tasks
On December 11, 2025, OpenAI presented the GPT-5.2 language model. The announcement coincided with the day Disney's partnership agreement was announced to integrate characters into its products. The model is positioned as optimized for professional use: creating tables, presentations, coding, image processing, working with long texts, and complex multi-step tasks.
What is known
In benchmarks, GPT-5.2 Thinking shows improved results compared to the previous version, GPT-5.1. In particular, in the AIME 2025 test, consisting of 30 complex mathematical problems, the model achieved 100% correct answers without using external tools, while GPT-5.1 showed results at the level of 94%. In the ARC-AGI-1 test, which checks abstract thinking ability, the result of GPT-5.2 exceeded the previous one by more than 10 percentage points.
According to an OpenAI statement, GPT-5.2 Thinking is 30% less likely to make errors in responses. The model is focused on use in research, analysis, text creation, and decision-making support. The updated GPT-5.2 Instant version has also been improved for everyday queries: instructions, technical writing, translations, and educational content.
OpenAI also notes that GPT-5.2 will be available in three versions: Instant, Thinking, and Pro. All options will initially be available to users with paid subscriptions. GPT-5.1 will remain available for three months in the deprecated models section.
Source: OpenAI
