December 30, 2025 – Yesterday, on December 29, tech media outlet BleepingComputer published a blog post revealing that Microsoft has initiated a full-scale rollout of the GPT 5.2 model for users of its web-based, Windows 11, and mobile Copilot services. This new iteration has been christened the “Smart Plus” mode.
This update comes as a complimentary enhancement, introducing the brand-new “Smart Plus” mode alongside the previously released “Smart” mode, which was based on GPT 5.1 and launched last month. The two modes will coexist, offering users a range of options.
The version of GPT 5.2 integrated into Copilot is specifically identified as the “Thinking” variant. Microsoft has placed a strong emphasis on its significantly improved ability to handle complex tasks.
This model is designed to empower users to complete a variety of real-world work tasks more efficiently. These include building spreadsheets, creating presentation slides, writing and reviewing code, comprehending lengthy documents, and processing images. Microsoft officially positions it as an “expert-level” tool capable of excelling in specific office-related tasks.
According to the blog post, during an evaluation of “knowledge-based work” tasks across 44 different professions using the GDPval assessment, the GPT-5.2 Thinking model outperformed or matched human industry experts in 70.9% of the test cases. In contrast, the previous GPT-5 model achieved this level of performance in only 38.8% of the cases.
Multiple authoritative benchmark tests have demonstrated the dominant technical prowess of GPT 5.2. In the software engineering test SWE-Bench Pro, the model scored 55.6%, while it achieved an impressive 80% in SWE-bench Verified. Both scores surpass those of the GPT-5.1 Thinking model.
Furthermore, GPT 5.2 achieved a perfect score of 100% in the AIME 2025 (American Invitational Mathematics Examination). It also scored 92.4% in the GPQA Diamond test, 88.7% in the CharXiv reasoning (including Python) test, and showed a substantial increase in its score in the ARC-AGI test.
