Google Updates Gemini 3.1 Pro: The Neural Network Has Improved in Reasoning and Animation Generation
Google has taken another step in the development of large language models — the company has presented an update for its flagship tool, Gemini 3.1 Pro. It is now much more effective at handling tasks that require not just extracting facts from a database, but building complex logical chains, deep analysis, and solving non-standard problems.
The effectiveness of the update was confirmed by the results of the ARC‑AGI‑2 benchmark. This test evaluates how capable an AI is at solving new logical puzzles, rather than just recognizing familiar patterns. While the previous version — Gemini 3 Pro — scored 31% on it, the new iteration achieved a result of 77%. This more than twofold increase indicates significant progress in the model's development.
Google emphasizes that Gemini 3.1 Pro is designed for cases where "a simple answer is not enough" — for example, for multi-stage research, synthesizing data from numerous disparate sources, or visualizing complex concepts.
In addition to enhancing analytical capabilities, the developers have also improved the model's creative functions. Particular attention was paid to generating animated graphics: now Gemini 3.1 Pro can create complex animated SVG images based on a text prompt.
The advantage of this approach lies not only in visual appeal but also in technological efficiency. Unlike video files, SVG animations are stored as compact code — they are not "heavy," retain sharpness, and do not lose quality even when significantly enlarged, for example, to the size of a billboard. This solution could be especially useful for web designers and developers: it can be used to enhance websites without sacrificing page loading speed.