Gemini 3 Flash is now obtainable in Gemini CLI, supporting high-frequency workflows widespread to terminal-based work. Gemini 3 Flash achieves a SWE-bench Verified rating of 78% for agentic coding, outperforming not solely the two.5 sequence, but in addition Gemini 3 Professional. Gemini 3 Flash was constructed to be extremely environment friendly, pushing the Pareto frontier of high quality vs. price and pace and is out there in preview at lower than 1 / 4 the price of Gemini 3 Professional. With two of our greatest fashions powering Gemini CLI, pace now not has to imply compromising high quality.
Begin utilizing Gemini 3 Flash with Gemini CLI
Beginning right this moment, most paid tier clients of Gemini CLI have entry to each Gemini 3 Professional and Gemini 3 Flash, together with:
- All non-business clients of Google AI Professional or AI Extremely
- Customers who’ve entry utilizing a paid API key by way of Google AI or Vertex
- Gemini Code Help customers which were enabled by their cloud admin for preview fashions
Free of charge tier customers:
- We’ve onboarded everybody who signed as much as the beforehand obtainable waitlist, so please examine your e-mail for particulars
- If you weren’t on our waitlist, we’re rolling out extra entry steadily to make sure the expertise stays quick and dependable, so keep tuned for extra particulars, or view our docs to find out about your choices for entry now
Get began by upgrading Gemini CLI model to the newest model (0.21.1):
npm set up -g @google/gemini-cli@newestPlain textual content
After you’ve confirmed your model is 0.21.1 or later, run /settings, then toggle the setting “Preview options” to true. When you’ve enabled preview options, run /mannequin to pick Gemini 3.
This launch brings the complete capabilities of the Gemini 3 household to your terminal. You’ll be able to depend on Gemini CLI’s clever auto-routing to order Gemini 3 Professional for extremely complicated reasoning, or use the handbook selector to dedicate a selected mannequin to your whole duties. The numerous reasoning enhancements in Gemini 3 Flash will let you execute prompts that beforehand required slower Professional-tier fashions, at a decrease price.
Construct something within the terminal with improved agentic coding
Gemini 3 Flash raises the efficiency flooring of your coding classes with sturdy efficiency in reasoning, device use, and multimodal capabilities.
Generate a ready-to-deploy app with 3D graphics
We used Gemini 3 Professional in Gemini CLI to construct a 3D Voxel simulation of the Golden Gate Bridge, treating the immediate as each a artistic temporary and a technical specification. However can Gemini 3 Flash do the identical?
Beforehand, producing this stage of purposeful code in a single go was a job extra suited to Professional fashions. Gemini 2.5 Flash, for instance, usually struggled with this complexity, leading to damaged logic. Whereas Gemini 3 Professional’s state-of-the-art reasoning creates a extra visually interesting outcome, Gemini 3 Flash can nonetheless deal with the duty with precision, demonstrating {that a} fast prototyping device would not need to compromise code high quality.
Enhance your day by day work
The true check of a improvement assistant is the way it handles the high-volume, sensible duties you execute all through the day. Gemini 3 Flash outperforms 2.5 Professional whereas being 3x quicker at a fraction of the associated fee (primarily based on Synthetic Evaluation benchmarking).
Motion code modifications from massive context home windows
Managing massive codebases usually includes sifting by way of a whole lot of feedback on a pull request to seek out the one actionable merchandise. This requires a mannequin able to holding an enormous context window with out shedding observe of particular directions.
On this demo, Gemini 3 Flash processes a simulated pull request thread containing 1,000 feedback. It efficiently cuts by way of pages of “bikeshedding” to find a single important request concerning a timeout adjustment. Gemini CLI then applies the exact replace to the configuration file on the primary strive. This demonstrates the mannequin’s capacity to tell apart sign from noise and execute correct edits inside large context home windows.
Simulate sensible consumer visitors for stress testing
Validating your backend infrastructure requires visitors that mimics precise consumer conduct, however writing customized load-testing scripts that deal with concurrency and particular consumer journeys is usually time consuming. A lot of these duties are nicely suited to Gemini 3 Flash, lowering syntax hallucinations and failure loops, whereas nonetheless offering quick responses.
On this demo, Gemini CLI is used to stress-test an internet utility hosted on Cloud Run. Gemini 3 Flash generates a Python script utilizing asyncio to simulate concurrent customers throughout three distinct situations: “Profitable Order,” “Cost Failed,” and “Stock Timeout.” When the preliminary execution returns protocol errors, the mannequin immediately analyzes the traceback and patches the script. This lets you launch a complete load check and observe the ensuing metrics in your Cloud Run dashboard in seconds.
Keep within the circulate longer
Gemini 3 Flash offers a brand new efficiency baseline for high-frequency improvement duties within the terminal. By elevating the efficiency flooring and integrating with Gemini CLI’s auto-routing, it goals that will help you work quicker and extra effectively. Whether or not you might be constructing a brand new prototype or managing complicated infrastructure, you now have a improvement assistant able to maintaining together with your tempo of labor.
Replace your Gemini CLI right this moment to the newest model to begin constructing quicker — at a decrease price per token — with Gemini 3 Flash.
