It’s a day of celebration for AI enthusiasts, engineers and developers when an announcement begins with the words, “Free and open source.”
Even more so when that announcement comes from a company like Google.
The company has released Gemini CLI, an open-source AI agent that integrates its Gemini 2.5 Pro model inside a user’s terminal. This allows users to execute commands and write code using natural, conversational prompts.
Google says it ensures users rarely hit a limit with its 60 free model requests per minute and 1,000 daily requests at no charge.
Besides coding, Gemini CLI can also handle various tasks within the terminal, including content creation, in-depth research, and task management.
While it might sound like Google is going after OpenAI’s Codex CLI and Anthropic’s Claude Code by shipping its own command-line tool, Gemini CLI might just add a whole new dimension to Google’s dominance.
Distribution is the True Moat
Two years ago, Emad Mostaque, the former CEO of Stability AI, stated that companies will not find an edge or a moat in AI due to talent, compute, or the powerful models they create, but rather in distribution.
Following the launch of Gemini CLI, Mostaque reshared his earlier 2023 comments, suggesting his prediction might be on track.
Not only has Google dominated the leaderboards in benchmark evaluations with their Gemini 2.5 Pro models, but it is also now moving toward a new paradigm in distribution.
“Gemini CLI is another example of how Google is going to drive the cost of generative AI down to $0,” Mostque said in reference to his prediction.
Moreover, it’s not just Gemini CLI. Google also offers free but limited access to its best AI models on the Gemini app. The $20 paid plans for Gemini are also bundled with Google Cloud storage.
Besides, the company features an array of free, open-source, lightweight models like the Gemma-3n, which not only offers excellent performance in its segment but can also be downloaded on Android devices for offline usage on the Edge Gallery app.
The list doesn’t end here. From tools like Notebook LM to AI features in Google Workspace, the company offers numerous free resources.
According to Mostaque, Google’s generosity can be attributed to the company’s ability to absorb losses, given its large-scale operations. Moreover, Google is also using its own chips (TPUs) to train and deploy AI models, instead of paying a premium for NVIDIA chips.
While Gemini CLI offers free usage, both Claude Code and Codex CLI are priced based on the number of tokens consumed.
“A sign of Google waking up is them offering by far and away the biggest free plans across the industry. This time is for the Claude Code competitor,” Nathan Lambert, a senior research scientist at the Allen Institute for AI, said.
While several social media users echoed Lambert’s sentiment, how does Google’s Gemini CLI compare to Claude Code?
Spoiler Alert: Claude Code May Just Be Better
Within 24 hours of Gemini CLI being launched, developers were quick to test it and compare it with Claude Code.
The latter has been in the news for its effectiveness in working wonders for several developers globally. Going by the comparison tests, it looks like Claude Code is winning—at least, for now.
Some developers claim that Anthropic’s flagship model, the Claude Opus 4, is more suitable for agentic tasks than Gemini 2.5 Pro. “It’s not even close right now,” Ian Nuttall, a developer and founder of numerous tech startups, said.
Meanwhile, Ryan Daigle, a developer, compared it with Claude Code and Devin AI on a medium-sized Elixir refactor task. Daigle did not mention any advantages for Gemini CLI besides being free with high limits and being “pretty fast” on each step.
He said there is no opportunity to review a plan before it is executed, unlike Claude and Devin. Moreover, he noted that it offers a very “jumpy terminal experience”.
“Claude code CLI is the king of the hill on TUI (terminal user interface),” he said.
He also noted a lack of mid-stream interactivity and said there’s no way to interrupt, add context, or redirect the agent. “Claude and Devin are great at taking instructions mid-stream,” he said.
Moreover, Diagle said he entered a ‘death loop’ on a relatively simple task and faced significant difficulty creating a pull request from local changes.
“Claude has a master of the ‘gh’ CLI and can do any git operation I give it. Devin speaks native git too,” he added.
Some other developers listed similar problems with their experience as well. Several users on Reddit also prefer Claude Code over Gemini CLI.
Having said that, several developers have also had positive experiences with Gemini CLI. One developer said on X that the free one-million-token context window helps feed the entire code base into the agent.
Yet, if Google ends up frequently updating Gemini CLI and follows its playbook of redemption, like it did with the Gemini family of models, how long will it take to catch up with Claude Code?
The post Is Google Driving the Cost of GenAI to $0? appeared first on Analytics India Magazine.
