Google Releases Gemini 3.1 Pro, Begins Gemini 4 Rollout
Google has launched Gemini 3.1 Pro, a new model focused on complex reasoning and multi-step problem solving, now available in preview for developers. The company is also in the process of a phased rollout for its Gemini 4 models, offering more granular API controls. Additionally, Google has introduced a new Gemini-powered tool for generating short music tracks.
- Gemini 3.1 Pro demonstrates a significant leap in reasoning capabilities over its predecessor, Gemini 3.0 Pro, more than doubling its score on the ARC-AGI-2 benchmark from 31.1% to 77.1%. - The model maintains a large context window of up to 1 million tokens, allowing it to process vast amounts of information, equivalent to roughly 1,500 pages of text. - Gemini 3.1 Pro introduces a new "medium" thinking level, designed to offer a balance between performance, speed, and cost that is comparable to the previous generation's "high" setting. - The new music generation tool, named Lyria 3, is an evolution of a model previously available to some YouTube creators and in Google's Music AI Sandbox for musicians. - Lyria 3 produces 30-second tracks for free users and can generate its own lyrics; it also uses a separate AI model called Nano Banana to create custom album art for each track. - To distinguish its outputs, all music generated by Lyria 3 in the Gemini app is embedded with a SynthID watermark, which marks the content as AI-generated. - While Gemini 4's official features are not yet detailed, expectations point toward improvements in "System 2" thinking, a slower, more deliberate form of reasoning aimed at reducing errors and improving logic in complex tasks. - Developers can access Gemini 3.1 Pro via AI Studio and Vertex AI, with pricing remaining the same as its predecessor at $2.00 per 1 million input tokens and $12.00 per 1 million output tokens.
