The company’s smartest model is now available on the terminal for Google AI Ultra subscribers and paid Gemini API key holders, with gradual rollout to ensure stability.
The new feature raises the level of reasoning, agentic coding and use of tools, allowing developers to perform complex tasks with natural commands.
The update arrives in version 0.16.x of the Gemini CLI and can be activated with the command npm install -g @google/gemini-cli@latest, followed by activating Preview features in the settings. Enterprise users of Gemini Code Assist will have access soon, while other plans can join the waitlist on the official GitHub.
Como ativar o Gemini 3 Pro imediatamente
Google AI Ultra subscribers access the model without additional waiting.
Simply update Gemini CLI to the latest version and enable preview features in the /settings menu.
The template becomes default automatically after activation.
Users with a paid Gemini API key also receive instant access.
The company recommends tracking rollout progress through specific discussion on GitHub.
Agentic coding creates complete applications in a single prompt
Gemini 3 Pro excels at synthesizing disparate information including text, images and code.
Developers can describe complex projects and receive functional scaffolds ready for execution.
A practical example is the creation of a photorealistic 3D simulation of the Golden Gate Bridge in a single HTML file, with Three.js via CDN, time controls, volumetric fog, instantiated traffic and optimization for 60 FPS.
The model generates all planning and implementation without the need for a build step, delivering a runnable application directly in Chrome.
Transformation of sketches into functional code with multimodal vision
Just drag a sketch image to the terminal and Gemini 3 Pro identifies layout, buttons and visual elements.
The system generates HTML, CSS and JavaScript ccomplete with precision.
A real example shown was a futuristic dark-mode interface for an internal brand intelligence tool, with glass pillars, luminous threads representing customer journeys and floating cards in Tailwind CSS.
Complex shell commands become natural language
Gemini 3 Pro translates intent into precise Unix commands, including advanced flags.
Runs tasks like git bisect automatically to find problematic commits.
Users describe the problem in natural language and receive the exact hash of the commit that changed settings, for example.
The model also interprets dense command output and summarizes it in clear language.
Professional documentation generated from existing code
Gemini 3 Pro analyzes the entire codebase before generating documentation.
Produces organized content, explaining user-visible features, command-line options, authentication, and technical concepts.
In open-source projects, includes architectural overview, component summary and contribution guide, maintaining consistent and searchable formatting.
Performance debugging in Cloud Run services with native integration
The model orchestrates multi-step flows connecting Cloud Run, Snyk, and version control.
Identifies the root cause of performance problems, suggests corrections and deploys automatically.
Turns investigation that required multiple tools into single action via Gemini CLI extensions.
- Build complete 3D prototypes in a single file
- Convert sketches into functional apps with drag-and-drop
- Run git bisect or complex commands by description
- Automatically generate complete and professional documentation
- Resolve performance problems in production safely
Rollout and future access
Google AI Ultra and paid API keys receive full priority.
Gemini Code Assist Enterprise gains access in the coming weeks.
Other users (Google AI Pro, Code Assist Standard and free tier) can join the official waiting list.
The company maintains public tracking on GitHub for progress transparency.
Gemini 3 Pro marks a significant leap in Terminal-Bench 2.0 (54.2%) and SWE-bench Verified (76.2%), consolidating leadership in agentic coding and multimodal reasoning directly in the terminal.