Google Releases Gemini 3.1 Pro Model
Google has launched Gemini 3.1 Pro, a new AI model focused on improved multimodal capabilities, efficiency, and lower cost-per-task. While benchmarks show strong performance, analysis suggests its success will depend on how well it integrates into existing developer workflows and competes on practical metrics like reliability and predictable cost, rather than just raw performance.
- On the ARC-AGI-2 benchmark, which measures the ability to solve novel logic patterns, Gemini 3.1 Pro scored 77.1%, more than double the reasoning performance of its predecessor, Gemini 3 Pro. It also outperforms competitors like Anthropic's Claude Opus 4.6 and OpenAI's GPT-5.2 on multiple key benchmarks, though OpenAI's GPT-5.3-Codex leads on the specialized SWE-Bench Pro coding test. - The model is priced at $2 per million input tokens and $12 per million output tokens, which is significantly less expensive than Claude Opus 4.6 and competitive with GPT-5.2. Despite having the same price as Gemini 3 Pro, version 3.1 demonstrates 10-15% lower output token consumption on average for identical tasks, offering a direct cost saving on usage. - Developers can access the preview of Gemini 3.1 Pro through the Gemini API in tools like Google AI Studio and Vertex AI, as well as through the Gemini CLI, Android Studio, and a new agentic development platform named Google Antigravity. - A novel feature for developers is the ability to generate website-ready, animated Scalable Vector Graphics (SVGs) directly from a text prompt, creating animations written in code rather than pixels. - This release follows a period of significant growth in developer adoption for the platform; between March and August 2025, the number of Gemini API calls grew by 143% from 35 billion to 85 billion. - Google's earlier Gemini models, like versions 1.0 and 1.5, operated at a negative margin to drive growth, while the release of Gemini 2.5 marked the first time the models achieved consistent positive margins by competing on quality rather than price. - The model introduces a new "Medium" thinking level via the API, allowing developers to better balance cost, performance, and speed for different tasks, such as using a lower level for simple autocomplete and a higher one for complex debugging. - Gemini 3.1 Pro is natively multimodal, designed to process text, images, audio, and video within a single prompt and features a context window of up to one million tokens.
