Google releases Gemini 3.1 Pro

Gemini 3.1 Pro is Google's new standard model. The new version is being integrated into the Gemini app and NotebookLM, as well as developer and enterprise tools. The model is expected to handle more complex tasks better than its predecessor.

Google sees the biggest difference to its predecessor in reasoning. 3.1 Pro is a smarter and more powerful model for processing complex problems. This finding is based on the ARC-AGI-2 benchmark, which requires logical patterns to be recognized. The new model delivers 77.1 percent correct solutions, while Gemini 3 Pro only achieved about half (31.1 percent).

In a blog post, Google shows a series of code-based visualizations and animations that can depict Gemini 3.1 Pro much more attractively and comprehensively. In addition to a moon, stars twinkle, and craters are visible on the surface. A moving flock of birds can be created via prompt and then even moved using hand tracking.

The agentic search (BrowseComp) is also noteworthy, with Gemini 3.1 Pro achieving 85.9 percent – compared to 59.2 percent for its predecessor. As always, the results are not necessarily transferable to one's own experiences and practical application. New models are trained on tests, among other things, so they perform particularly well here. Nevertheless, Gemini 3 had also proven to be a particularly popular model that is said to have even put OpenAI under pressure.

OpenAI recently released models primarily for developers – Codex 5.3 along with a new platform. Anthropic had also previously made Claude Opus 4.6 available, which comes with improved coding capabilities. Most recently, the mid-tier model of the Claude family, Sonnet, was also upgraded to a new version (4.6).

(emw)