Google Launches Gemini 3.1 Pro and Gemini 2.5 Computer Use — AI That Controls Your Browser Joins the Agent Race Against OpenAI and Anthropic

Google DeepMind and the Google Gemini team officially released two significant AI capabilities in March 2026 that collectively advance Google’s position in the rapidly expanding AI agent market. The first is Gemini 3.1 Pro — Google’s most capable reasoning model, available via Google AI Studio, Vertex AI, and the Gemini API. Google CEO Sundar Pichai confirmed in a public statement that the updated model addresses feedback from earlier versions with improvements to its style, structure, and creative formatting, enabling more nuanced and better-formatted responses for enterprise and developer use cases. The second is Gemini 2.5 Computer Use — a new model specifically designed to control web browsers autonomously, allowing developers to build agents that navigate websites, interact with web applications, fill forms, click elements, and complete multi-step web tasks just as a human user would. Gemini 2.5 Computer Use is accessible through Google AI Studio and Vertex AI, with a public demonstration hosted on Browserbase allowing anyone to observe the AI completing tasks including browsing Hacker News and playing browser-based games. Google’s approach deliberately focuses on browser-level interactions rather than full desktop or operating system control — a design choice Google acknowledges is intentional while the technology matures. Google’s official blog states the model ‘outperforms leading alternatives on multiple web and mobile benchmarks.’ The launch positions Google alongside Anthropic — which released Claude with computer use capabilities in October 2025 — and OpenAI, which offers its ChatGPT Agent, in a three-way race to dominate the AI agent automation market. Simultaneously, Anthropic and OpenAI have both released smaller, cheaper models for high-volume developer workloads: Claude Haiku 4.5 at $1 per million input tokens, and GPT-5.4 nano at $0.20 per million input tokens — with Google’s Gemini 2.5 Flash-Lite offering the lowest published price at $0.10 per million input tokens.