Google has officially announced the release of its new flagship artificial intelligence model — Gemini 3.1 Pro. According to the developers, this version is smarter and more functional than its predecessor and is now available for review.
This is reported by Business • Media
Comparison with Previous Models and Performance
Gemini 3.1 Pro is an updated version of Gemini 3 Pro, which was introduced last November. Both models are based on the MoE architecture; however, the new version shows significant advantages. In the Humanity’s Last Exam test, which assesses knowledge depth across various fields, Gemini 3.1 Pro achieved a score of 44.4%, while Gemini 3 Pro scored 37.5%, and GPT-5.2 scored 34.5%.
According to experts from Artificial Analysis, the new model leads in six out of ten major benchmarks and is noted for much more efficient token usage compared to other flagship solutions. Meanwhile, the model’s architecture remains similar to the previous version, but optimizations focus on deeper analysis, response stability, and resource cost efficiency.
“In particular, in the popular Humanity’s Last Exam test, which examines knowledge depth in specific fields, the new model scored 44.4%. Gemini 3 Pro scored 37.5%, and GPT-5.2 scored 34.5%.”
New Features and Expanded Capabilities of Gemini 3.1 Pro
The updated Gemini 3.1 Pro supports working with text, audio, video, images, and PDF files. It can process over 1 million input tokens and generate over 65,000 output tokens. In the ARC-AGI-2 benchmark, the model achieved a score of 77.1%, more than double that of Gemini 3 Pro.
Among the new features are the generation of scalable SVG animations based on text prompts, the creation of interactive dashboards for complex systems, and the development of 3D simulations with real-time hand tracking and sound generation. The model can also transform abstract literary themes into visual interfaces, such as the novel “Wuthering Heights,” where Gemini 3.1 Pro created an interface that conveys the atmosphere of the work.
Despite significant progress, users have noted some shortcomings. In particular, the community reports that the model tends to get stuck in certain cases, similar to previous versions in this line.
Starting February 19, 2026, Gemini 3.1 Pro will be available to developers, enterprises, and regular users. Enhanced token usage limits will apply for subscribers of Google AI Pro and Ultra.