Forem: Taylor Mullen

Gemini 3 Flash is now available in Gemini CLI

Taylor Mullen — Wed, 17 Dec 2025 18:41:52 +0000

Gemini 3 Flash is now available in Gemini CLI, supporting high-frequency workflows common to terminal-based work. Gemini 3 Flash achieves a SWE-bench Verified score of 78% for agentic coding, outperforming not only the 2.5 series, but also Gemini 3 Pro. Gemini 3 Flash was built to be highly efficient, pushing the Pareto frontier of quality vs. cost and speed and is available in preview at less than a quarter the cost of Gemini 3 Pro. With two of our best models powering Gemini CLI, speed no longer has to mean compromising quality.

Start using Gemini 3 Flash with Gemini CLI

Starting today, most paid tier customers of Gemini CLI have access to both Gemini 3 Pro and Gemini 3 Flash, including:

All non-business customers of Google AI Pro or AI Ultra
Users who have access using a paid API key through Google AI or Vertex
Gemini Code Assist users that have been enabled by their cloud admin for preview models
For free tier users:
We’ve onboarded everyone who signed up to the previously available waitlist, so please check your email for details
If you were not on our waitlist, we’re rolling out additional access gradually to ensure the experience remains fast and reliable, so stay tuned for more details, or view our docs to learn about your options for access now
Get started by upgrading Gemini CLI version to the latest version (0.21.1):

npm install -g @google/gemini-cli@latest

After you’ve confirmed your version is 0.21.1 or later, run /settings, then toggle the setting “Preview features” to true. Once you’ve enabled preview features, run /model to select Gemini 3.

This release brings the full capabilities of the Gemini 3 family to your terminal. You can rely on Gemini CLI’s intelligent auto-routing to reserve Gemini 3 Pro for highly complex reasoning, or use the manual selector to dedicate a specific model to all of your tasks. The significant reasoning improvements in Gemini 3 Flash allow you to execute prompts that previously required slower Pro-tier models, at a lower cost.

Build anything in the terminal with improved agentic coding

Gemini 3 Flash raises the performance floor of your coding sessions with strong performance in reasoning, tool use, and multimodal capabilities.

Generate a ready-to-deploy app with 3D graphics

We used Gemini 3 Pro in Gemini CLI to build a 3D Voxel simulation of the Golden Gate Bridge, treating the prompt as both a creative brief and a technical specification. But can Gemini 3 Flash do the same?

Previously, generating this level of functional code in a single pass was a job more suited for Pro models. Gemini 2.5 Flash, for example, often struggled with this complexity, resulting in broken logic. While Gemini 3 Pro's state-of-the-art reasoning creates a more visually appealing result, Gemini 3 Flash can still handle the task with precision, demonstrating that a rapid prototyping tool doesn't have to compromise code quality.

Improve your daily work

The true test of a development assistant is how it handles the high-volume, practical tasks you execute throughout the day. Gemini 3 Flash outperforms 2.5 Pro while being 3x faster at a fraction of the cost (based on Artificial Analysis benchmarking).

Action code changes from large context windows

Managing large codebases often involves sifting through hundreds of comments on a pull request to find the single actionable item. This requires a model capable of holding a massive context window without losing track of specific instructions.

In this demo, Gemini 3 Flash processes a simulated pull request thread containing 1,000 comments. It successfully cuts through pages of "bikeshedding" to locate a single critical request regarding a timeout adjustment. Gemini CLI then applies the precise update to the configuration file on the first try. This demonstrates the model’s ability to distinguish signal from noise and execute accurate edits within massive context windows.

Simulate realistic user traffic for stress testing

Validating your backend infrastructure requires traffic that mimics actual user behavior, but writing custom load-testing scripts that handle concurrency and specific user journeys is often time consuming. These types of tasks are well suited for Gemini 3 Flash, reducing syntax hallucinations and failure loops, while still providing fast responses.

In this demo, Gemini CLI is used to stress-test a web application hosted on Cloud Run. Gemini 3 Flash generates a Python script using asyncio to simulate concurrent users across three distinct scenarios: "Successful Order," "Payment Failed," and "Inventory Timeout." When the initial execution returns protocol errors, the model instantly analyzes the traceback and patches the script. This allows you to launch a comprehensive load test and observe the resulting metrics in your Cloud Run dashboard in seconds.

Stay in the flow longer

Gemini 3 Flash provides a new performance baseline for high-frequency development tasks in the terminal. By raising the performance floor and integrating with Gemini CLI’s auto-routing, it aims to help you work faster and more efficiently. Whether you are building a new prototype or managing complex infrastructure, you now have a development assistant capable of keeping up with your pace of work.

Update your Gemini CLI today to the latest version to start building faster — at a lower cost per token — with Gemini 3 Flash.

5 things to try with Gemini 3 Pro in Gemini CLI

Taylor Mullen — Tue, 18 Nov 2025 19:01:52 +0000

Gemini 3 Pro is now available in Gemini CLI

We've integrated Gemini 3 Pro, our most intelligent model, directly into Gemini CLI to unlock a new level of performance and productivity in the terminal. This powerful combination delivers state-of-the-art reasoning for executing better commands, enhances support for complex engineering work through agentic coding, and enables smarter, more tailored workflows via advanced tool use.

We are rolling out access gradually to ensure the experience remains fast and reliable.

Gemini 3 Pro is available starting now in Gemini CLI for Google AI Ultra subscribers (Google AI Ultra for Business access is on the roadmap) and for those who have access via paid Gemini API key.
For Gemini Code Assist Enterprise users, access is coming soon.
All other users, including Google AI Pro, Gemini Code Assist Standard, and free tier users can join the waitlist here to get access as it becomes available.

You can also track our rollout progress by following this GitHub discussion.

Start using Gemini 3 Pro with Gemini CLI

If you’re a Google AI Ultra subscriber or have a paid Gemini API key, get started immediately by upgrading your Gemini CLI version to 0.16.x:

npm install -g @google/gemini-cli@latest

After you’ve confirmed your version, run /settings, then toggle Preview features to true. Gemini CLI will now default to Gemini 3 Pro.

Users can follow the instructions in this video to learn how to enable Gemini 3 Pro in Gemini CLI

Here are 5 practical ways you can tap into Gemini 3 Pro in Gemini CLI to accelerate development and bring your biggest ideas to life.

Build anything in the terminal with improved agentic coding

Gemini 3 Pro excels at coding because of its ability to synthesize disparate pieces of information, including text, images, and code, and follow complex, creative instructions. It understands the intent behind your idea, allowing you to go from a rough concept to a functional starting point in a single step.

1. Generate a ready-to-deploy app with 3D graphics
Gemini 3 Pro's agentic coding capability allows it to handle prompts that are both a creative brief and a technical spec at the same time. It can take a prompt, create a detailed execution plan, and then generate the entire scaffold for a runnable web project, not just a single file.

For example, say you have an idea for a visually impressive prototype—a 3D graphic for a landing page or a quick tech demo. Instead of spending hours setting up a graphics library and a local dev server, you can describe the entire project in one go and get a working starting point immediately.

Objective: Build a visually stunning, photorealistic 3D Voxel simulation of the Golden Gate Bridge using Three.js, prioritizing quality and complex visuals (no simple blocks),  atmospheric depth and 60FPS performance.

Visuals & Atmosphere:
- Lighting: Slider (0-24h) controlling sun position, light intensity, sky color, and fog color.
- Fog: Volumetric-style fog using sprite particles that drift and bob. Slider 0-100. 0 = True Zero (Crystal Clear). 100 = Dense but realistic (not whiteout).
- Water: Custom GLSL shader with waves, specular reflections, and manual distance-based fog blending (exp2) for seamless horizon integration.
- Post-Processing: ACESFilmic Tone Mapping and UnrealBloom (optimized for glowing lights at night).

Scene Details:
- Bridge: Art Deco towers with concrete piers (anchored to seabed), main span catenary cables, and suspenders.
- Terrain: Low-poly Marin Headlands and SF Peninsula.
- Skyline: Procedural city blocks on the SF side.
- Traffic: Up to 400 cars using `InstancedMesh`, positioned accurately on top of the deck (ensure vertical alignment prevents clipping into the concrete). Each car features emissive headlights (white) and taillights (red).
- Ships: Procedural cargo ships with hull, containers, and functional navigation lights (Port/Starboard/Mast/Cabin) moving along the water.
- Nature: Animated flocking birds.
- Night Mode: At night, activate city lights, car headlights, ship navigation lights, tower beacons, street lights.

Tech & Controls:
- Core: Must output only single HTML file `golden_gate_bridge.html` to be run in a blank Chrome tab. Import Three.js/Addons via CDN map.
  -   `three` (Core library) via CDN (ES Modules).
  -   `three/examples/jsm/...` modules via Import Map.
  -   No build step (Vite/Webpack). Pure HTML/JS.

- UI: Visually appealing sliders for Time (0-24h), Fog Density (0-100%), Traffic Density (0-100%), and Camera Zoom.
- Optimization: `InstancedMesh` for all repetitive elements (cars, lights, birds).

2. Turn a visual idea into a working app

You've sketched a UI and need to translate that visual concept into functional code. You can take a picture of your sketch, and then simply drag and drop the image file into your terminal.

Gemini 3 Pro's multimodal understanding will analyze the drawing, identifying buttons, text boxes, and layout. It then generates the HTML, CSS, and JavaScript code to bring your sketch to life.

Create a UI for "Project Constellation," an internal brand intelligence tool prototype that shows a customer acquisition pipeline. The aesthetic is an ultra-creative, futuristic dark-mode nebula. Luminous, iridescent threads representing customer journeys weave through semi-transparent glass pillars. A sleek, floating data card with Tailwind CSS precision materializes when hovering over a pillar. I've prepared a sketch for you to work from: @sketch.png.

Improve your daily work

While vibe coding demos show off the art of the possible, the true test of a developer tool is how it performs the practical, everyday work you do multiple times a day. Small improvements in these common workflows, like refactoring code, debugging errors, or managing infrastructure, are what create real productivity gains.

This is where Gemini 3 Pro's state-of-the-art reasoning makes a tangible difference. It follows the nuances of complex, multi-part commands with greater precision than ever before, which is essential for the practical, detailed work that defines your day.

Here are a few examples of how Gemini 3 Pro can handle these critical engineering tasks.

3. Generate complex shell commands with natural language

With Gemini CLI, the power of the UNIX command line is available directly through natural language. No need to memorize the obscure syntax and every flag of UNIX commands, simply have Gemini 3 Pro translate your intent and execute it for you. Gemini can then even parse dense formatted output back into natural language for you.

Ask Gemini CLI to handle all the complexity of running Git Bisect for you on the command line, leaving you free to focus on applying your judgement on finding the bug in question.

At some point I lost the commit that set my default theme to dark. 
Find it for me with git bisect and return the hash to me.

4. Generate accurate documentation from your code

Gemini 3 Pro's advanced reasoning allows it to read and understand the logic of your codebase. It doesn't just see syntax; it can investigate and synthesize the purpose of a function, identify its parameters and return values, and translate that complex logic into clear, human-readable language.

This is useful when you’ve introduced a complex application and now need to create the documentation. Instead of manually writing out descriptions, you can have Gemini analyze the code and generate the docs for you in a format that is consistent with your code.

"This is an application that does not have any documentation and we do not have a technical writer. Before you begin, review all of the code. Then make me a user documentation. This document should only explain user facing features, but make sure to explain every single feature such as usage of the app, command line options, authentication options, built in tools, and all other user facing features. For certain features such as MCP or extensions, also explain the topic and concept so that the user has a better understanding. Since this is an open source project, provide an architectural overview of how the code is laid out, a summary of each component, and how they can contribute to the open-source project. The document should be organized and formatted so that it is easy to read and find. Do not make it a single html page. Make sure to add a search feature."

5. Debug performance issue in a live Cloud Run service

Gemini 3 Pro can orchestrate complex workflows across different services that hold your team's context. The improved tool use means it can plan and execute multi-step tasks that require gathering information from several sources—like observability, security, and source control—to solve a single problem.

In this example it connects a serverless platform (Cloud Run) with a popular security scanner (Snyk) using Gemini CLI extensions to find the root cause and suggest a fix, then deploys the fix, turning a complex, multi-tool investigation into a single, streamlined action.

Users are reporting that the "Save Changes" button is slow, investigate the 'tech-stack' service

Learn more today

These examples are just the start. The real potential isn't in running these specific commands, but in how Gemini 3 Pro can adapt to your unique challenges whether you're optimizing daily shell commands, tackling substantial engineering work, or building a workflow personalized to your team's tools. Gemini 3 Pro transforms the command line into an intelligent partner that understands your context.

The best way to see the difference is to try it yourself. Visit the Gemini CLI website, and share your own examples on social with #GeminiCLI. We can't wait to see what you build.