Just one week after the release of Gemini 3 Deep Think, Google is upping the ante again. The new Gemini 3.1 Pro model is now being positioned as the central intelligence for complex workflows. While its immediate predecessor already delivered solid performance, version 3.1 is expected to represent a significant leap forward, particularly in logical reasoning and direct collaboration with autonomous agent platforms like Google Antigravity.
Improved reasoning performance in the ARC-AGI-2 benchmark
The technical focus of Gemini 3.1 Pro is heavily on increasing so-called reasoning capabilities. According to Google This was deliberately done by the ARC-AGI-2 Benchmark validated, which tests a model's ability to solve entirely new logic patterns instead of simply retrieving known information from training data. With a verified score of 77,1 percent, the new model nearly doubles the performance of Gemini 3 Pro in this specific area. The context window remains open with up to... 2 million tokens The usual large size allows for the analysis of extremely large datasets or hours of video recordings.
Software integration: Antigravity and NotebookLM
The model is being distributed in two ways, for developers and end users. Of particular relevance to the coding community is its deep integration into the new agent platform. Google AntigravityThis is a specialized environment in which the AI no longer just completes code fragments, but as an autonomous actor takes over complete task chains including testing and deployment.
In the consumer sector, the update is particularly noticeable in NotebookLM Noticeable. Here, Google is using its increased logic power to present complex scientific topics in a more structured way. A new feature also allows animated SVGs to be generated directly from text prompts. This makes it possible to explain concepts visually using vector-based code instead of simple pixel graphics.
Pricing, subscriptions and availability
The model will initially launch as a preview version to gather practical feedback. However, availability is strictly tied to specific subscription plans. Access to the regular Gemini app and NotebookLM remains exclusively for paying users with a [specific subscription type]. Google AI Pro or Ultra subscription Reserved.
Developers and businesses access the model via the Gemini API in Google AI Studio, the Gemini CLI, or Android Studio. The enterprise pricing structure via Vertex AI continues to be based on the number of tokens used, with higher fees for queries requiring particularly large context (over 200.000 tokens). Ultimately, with version 3.1 Pro, Google solidifies its claim to provide the infrastructure for the next generation of autonomous AI agents.