Gemini 3 Flash is now available in Gemini CLI
Source: Google Developers Blog
Overview
Dec 17, 2025 – 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 of 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 via a paid API key through Google AI or Vertex
- Gemini Code Assist users that have been enabled by their cloud admin for preview models
(see admin instructions)
For free‑tier users
- Everyone who signed up on the previous waitlist has been onboarded – check your email for details.
- If you were not on the waitlist, we’re rolling out additional access gradually to keep the experience fast and reliable. Stay tuned, or view our docs to learn about your current options.
Upgrade Gemini CLI
Upgrade to the latest version (0.21.1 or newer):
npm install -g @google/gemini-cli@latestConfirm your version is 0.21.1 or later, then:
- Run
/settingsand toggle Preview features to true. - Run
/modeland select Gemini 3.
Model selector
What’s new?
This release brings the full capabilities of the Gemini 3 family to your terminal:
- Intelligent auto‑routing: Gemini CLI automatically reserves Gemini 3 Pro for highly complex reasoning tasks.
- Manual selector: Choose a specific model for all of your tasks if you prefer.
- Gemini 3 Flash: Significant reasoning improvements let you run prompts that previously required the slower Pro tier, now 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 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.
Video placeholder – your browser may not support playback.
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 3× 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.
Video unavailable: your browser doesn’t support playback for this video.
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 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
- 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.
Video unavailable: your browser doesn’t support playback for this video.
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 helps you work faster and more efficiently. Whether you’re building a new prototype or managing complex infrastructure, you now have a development assistant that can keep 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.