Gemini CLI 6.0: Google JUST INTEGRATED Gemini-3 into Gemini CLI & IT'S INSANE!

By AICodeKing

AI Command-Line InterfacesAI Model IntegrationSoftware Development ToolsUser Interface Enhancements
Share:

Key Concepts

  • Gemini 3 Pro
  • Gemini CLI
  • Waitlist
  • API Key
  • npm install
  • Mouse-based navigation
  • Sticky headers
  • Prompt box anchoring
  • Preview features
  • /models command
  • /stats command
  • TPU chips
  • Inference
  • Expo
  • Go-based terminal calculator
  • Godot game
  • Kanban app (Svelte)
  • Nuxt app
  • Tauri image cropper
  • Open code repo
  • Agentic contraption
  • Fallback to Gemini 2.5 Pro

Gemini CLI Integration with Gemini 3

Google is progressively integrating Gemini 3 into its suite of tools, with the Gemini CLI being the latest to receive this update. While direct Gemini 3 access in the CLI currently requires joining a waitlist, the process is described as responsive, with over 80% of applications already processed and expected access within a day or two for new sign-ups. Alternatively, users can leverage the API by setting up an API key for immediate access without the waitlist.

Enhancements to the Gemini CLI Interface

Significant improvements have been made to the Gemini CLI's foundation, aiming to reduce visual clutter and enhance user experience. Key interface updates include:

  • Rewritten Foundation: The rendering engine has been overhauled to eliminate visual noise and provide a polished, graphical-like interface within the terminal.
  • Mouse-based Navigation: Users can now navigate and interact with the CLI using a mouse, facilitating easier text copying and pasting.
  • Sticky Headers: These headers ensure that the currently edited file remains visible even when scrolling through multiple files.
  • Resizing Stability: The terminal no longer flickers or exhibits visual glitches when resized.
  • Anchored Prompt Box: The input field is now consistently positioned at the bottom of the terminal for a predictable user experience.

Accessing and Configuring Gemini 3 in the CLI

To utilize Gemini 3 within the CLI, users who have been accepted from the waitlist need to perform a few configuration steps:

  1. Run Settings Command: Execute the settings command to open the configuration interface.
  2. Enable Preview Features: Within the settings, enable the "preview features" option, which is crucial for accessing the new models.
  3. Select Model: After enabling preview features, run the /models command. This will present a dropdown menu allowing users to select their desired model, such as Gemini 3 Pro or Gemini 2.5 Flash. Gemini 3 Pro is recommended as the superior option.

Subscription tiers (Ultra, Pro, Free) will influence usage limits, with Ultra subscribers receiving better allowances.

Model Usage and Observations

It has been noted that the CLI may sometimes fall back to Gemini 2.5 Pro, particularly during periods of high load. The absence of a visible indicator showing which model is currently in use is a point of concern for the presenter. However, the /stats command with the model subcommand can be used to view the models utilized after a task has been completed.

Performance and Benchmarking with Gemini 3

Gemini 3 in the CLI demonstrates impressive speed, attributed to Google's use of their own TPU chips for both training and inference, which contributes to high speeds and potentially lower API costs. Several benchmark tests were conducted:

  • Movie Tracker App (Expo): Gemini CLI generated a functional movie tracker app using Expo, described as "quite good."
  • Go-based Terminal Calculator: The generated calculator is "really good," "awesome," and resembles an iOS calculator.
  • Godot Game (FPS): The CLI successfully edited an FPS game to include a step calculator and a health bar affected by jumping. While generations from other models like Kilo, Klein, or Rue might be superior, this result is considered "not bad."
  • Kanban App (Svelte): This generation is highlighted as one of the best, surpassing Kilo or Rue's output and being "really good to use" and deployable in a single step.

However, some tasks did not perform as expected:

  • Nuxt App: Did not work.
  • Tauri Image Cropper: Did not work.
  • Open Code Repo: Did not work.

These failures are attributed to the fallback mechanism to Gemini 2.5 Pro for long-running tasks, which is considered a limitation for agentic functionalities. The presenter hopes for better load balancing and limit management as the rollout progresses.

Conclusion and Future Hopes

The presenter expresses strong satisfaction with the advancements in Gemini CLI, particularly the improved UI and stability. Key hopes for future development include clearer limits for Gemini 3 and the implementation of a feature to display the exact model being used at any given time, rather than automatic switching. Overall, the Gemini CLI with Gemini 3 integration is considered a significant and positive development.

Chat with this Video

AI-Powered

Hi! I can answer questions about this video "Gemini CLI 6.0: Google JUST INTEGRATED Gemini-3 into Gemini CLI & IT'S INSANE!". What would you like to know?

Chat is based on the transcript of this video and may not be 100% accurate.

Related Videos

Ready to summarize another video?

Summarize YouTube Video