Google's top AI model Gemini 2.5 Pro free and open source for the console
Google's top model, Gemini 2.5 Pro, performs common developer and admin tasks on the console, including deployment, independently and free of charge.
Google Gemini-App.
(Image: mundissima/Shutterstock.com)
With Gemini CLI, Google has released a coding and admin assistant for the command line that has access to all system functions that can be controlled with commands: Commands for the operating system, console tools as well as system and network resources. Gemini CLI is based on Google's top model Gemini 2.5 Pro and is available under an Apache 2 license, whereby it is used via the cloud and is free of charge for common developer tasks.
Developers operate the console AI with prompts in natural language and complete programming tasks, including testing and deployment. The tool has access to repositories and pipelines. It clones dependencies independently, analyzes the tasks, asks questions and detects its errors. It integrates resources such as PDF documents or multimedia content and uses external sources such as Google searches via the Model Context Protocol (MCP).
For administrative tasks, in addition to the complete set of terminal commands, it masters function calling or the help button --help, which most console tools offer to open up the use of the tools. Users retain control over the history and decide at each step whether Gemini should execute it – once, permanently or not at all. There is also a fully automated, non-interactive mode.
Videos by heise
Containers and VS Code
Google publishes versions for Linux, Mac and Windows, but the assistant also runs in containers and via Code Assist in VS Code, Slack, Teams and other tools. Google recommends using it in containers for security reasons, as the AI can intervene heavily in the system.
The tool is written in Go, is licensed under Apache 2 and its behavior can be adapted via markdown files in the code base. The manufacturer has not planned a locally running version; the Gemini 2.5 Pro model itself works entirely in the cloud.
One instance with a context of up to one million tokens, 60 requests per minute and 1000 per day is free of charge, which should be sufficient for all common use cases in day-to-day development. Users who require multiple instances use a key from AI Studio or Vertex and pay according to the plan provided there.
(who)
