Google pushes Gemini 3.1 Pro as the new Pro baseline

Google is rolling out Gemini 3.1 Pro, calling it the upgraded core intelligence behind last week’s Gemini 3 Deep Think update, built for work where a quick reply usually isn’t enough, and keep reasoning on track.

It is launching across Google’s stack at once. Developers can access it in preview through the Gemini API in AI Studio and across other platforms. Consumers get it in the app, with higher limits for Google AI Pro and Ultra, and NotebookLM access is highlighted for Pro and Ultra users. The timing also lands right after Gemini added music generation powered by DeepMind’s Lyria 3, a separate beta feature that turns prompts or images into 30-second tracks.

Google’s headline proof point is ARC-AGI-2. It says Gemini 3.1 Pro scored a verified 77.1 percent on the benchmark, which is meant to test whether a model can solve new logic patterns, and it claims that is more than double Gemini 3 Pro’s reasoning performance.

The preview model is listed as gemini-3.1-pro-preview with a 1M input context window and 64k output, and a January 2025 knowledge cutoff. It supports long-context workflows and tool-heavy work with features like structured outputs and function calling. The same documentation also makes the boundaries clear, including no image generation and no audio generation in this model endpoint.

Google lists $2 per million input tokens and $12 per million output tokens up to 200k input tokens, with higher rates above that threshold, plus separate costs for context caching and Google Search grounding after a free monthly allowance.

Google says it is releasing 3.1 Pro in preview to validate the updates and keep improving agent-style workflows before it becomes generally available, so the practical question is whether it stays steady when the task looks less like chatting and more like finishing a full project.