On May 19, 2026, at the same hour a developer named Andrea Alberti watched a 27-commit pull request get merged into Gemini CLI, Google announced that the open-source AI coding tool would stop serving free users on June 18 — and that its replacement would not be open source at all. Alberti's question, posted minutes later in the project's GitHub discussion thread, put into plain language what hundreds of developers were suddenly asking: were they "essentially working for free on a code base that will only be used in enterprises?"
Google had accepted more than 6,000 merged pull requests from external contributors over nearly a year, cited those contributions as evidence of the project's success, and then announced that the community which supplied that labor would lose access while paying enterprise customers kept everything. The reaction was immediate. The announcement received 31 thumbs-down reactions on GitHub — the top reaction on the post. Developer @anthuanvasquez summarized the mood in two words: "As always, Google being Google."
Google's Pattern With Open Source: Attract, Extract, Close
The specific mechanics of what Google did are worth tracing precisely, because they differ from a straightforward product cancellation.
Google released Gemini CLI under the Apache 2.0 license in June 2025 — a permissive license that invites contribution, modification, and redistribution. The project grew quickly. By the time of the May 19 announcement, it had accumulated over 104,000 GitHub stars and more than 6,000 merged pull requests from external contributors. Google's own announcement cited those numbers proudly, calling them evidence that the community "love[s] a good terminal UI" and that developer "workflows have simply outgrown those early days of 2025."
What the announcement then did with that evidence was notable. Rather than treating community investment as a reason to maintain the open-source commitment, Google used it as justification for the transition: the project had proven itself, and now it was time to consolidate it into a proprietary platform. Starting June 18, free users, Google AI Pro and Ultra subscribers, and individual Gemini Code Assist users will stop receiving API responses. Enterprise customers with paid Standard or Enterprise licenses will retain access — and keep receiving model updates and security fixes.
Contributor @lingyaochu named the pattern directly in the discussion thread: "That's what you done, Google, open source Gemini CLI, get developers contribute to this, and then migrate the code to a close source project."
Christine Hall, a journalist who has covered FOSS since 2002, published her analysis at FOSS Force on May 22, describing it as "a new twist on a game that's become increasingly common in recent years." Her assessment: Google's action "effectively removes only the open source part of the equation" — not by changing the license, but by switching off the infrastructure that made the tool functional for anyone not paying. The sunsetting of Gemini CLI, she noted, "only affects open source users. It doesn't happen for those who pay for the product."
What the Apache License Does Not Protect
Google product manager Dmitry Lyalin addressed the open-source question in the GitHub announcement: "The project remains available to the community as an Apache 2.0 licensed repository with no changes."
This is technically accurate. It is also, contributors argue, deliberately incomplete.
An Apache 2.0 license governs code. It grants rights to use, modify, and redistribute source files. It says nothing about API access, authentication tokens, model quotas, or the cloud backend that determines whether a request receives a response. Simon Willison, the creator of Datasette and co-creator of the Django web framework, noted on Hacker News what developers discovered within hours of the announcement: the Antigravity CLI GitHub repository — introduced as the replacement — contains a changelog, a README, and an animated GIF demo. No source code.
Contributor @secretnarwhal reduced the announcement to its operational meaning: "So, basically, we're making the project closed source. That's it. That's the whole announcement."
@N1kky-wed went further: "What was the point of thousands of community contributions if the end result is a closed source replacement we can't meaningfully benefit from or contribute to? A lot of us spent time fixing bugs, building features, and helping grow Gemini CLI because we believed in the open ecosystem around it. Now it feels like the community helped validate the product direction, only for the future platform to move behind closed doors."
Google has not replied to those questions in the discussion thread.
Antigravity CLI Arrived Incomplete and More Expensive to Use
The Antigravity CLI that developers are being asked to migrate to by June 18 compounded the frustration. Built in Go rather than TypeScript, it supports asynchronous multi-agent workflows — a genuine technical improvement. In practice, users found the replacement unready.
As of two days after launch, the agy binary was not available on npm or Homebrew as a named package. Google's own announcement acknowledged the tool "won't have 1:1 feature parity right out of the gate." Multiple users on the Antigravity CLI issues tracker reported exhausting their weekly usage quota within a small number of requests — in some cases fewer than 2,000 lines of generated code. Gemini CLI had offered 1,000 requests per day; Antigravity CLI's quota refreshes weekly, not daily.
Developer @srameshc wrote that the chaos had eroded trust in Google's product management entirely: "I am loosing my faith in these managers and Google." On Hacker News, developer mpalmer put the cost plainly: "Say goodbye to metered usage via API keys you control, and hello to opaque pricing and usage limits."
Google's Defense Contradicts Itself
Google's stated rationale for the transition — a need for a "unified platform" to support multi-agent workflows — would be more persuasive if it applied to everyone. It does not.
Enterprise customers with paid Gemini Code Assist Standard or Enterprise licenses are explicitly exempted from the forced migration. They retain Gemini CLI, model updates, security patches, and can optionally adopt Antigravity CLI alongside it. The "technical imperative" that requires free users to abandon a tool they helped build turns out to be entirely optional for the customers Google most values financially.
That asymmetry is what contributors find hardest to accept. Developer @Neinndall, testing Antigravity CLI the same day, summarized the community's judgment: "Honestly do better."
Linux Foundation Tool Scores Exactly What Google Exploited
The episode has given fresh urgency to an instrument released specifically to catch this kind of claim before developers invest in it.
The Linux Foundation's Model Openness Tool, accessible at isitopen.ai, scores any AI model or tooling project across 17 components of the model development lifecycle — including model weights, training data availability, architecture documentation, fine-tuning scripts, inference code, and licensing terms. It produces a classification of closed (Class 1), partially open (Class 2), or fully open (Class 3). The framework it operationalizes — the Model Openness Framework, developed with researchers from the University of Oxford, Columbia University, and IBM — was built to counter what its authors called the practice of "claiming to be open source" while keeping critical components restricted.
Arnaud Le Hors of IBM, who co-authored the framework and presented it at Open Source Summit North America 2026 in Minneapolis this month, explained the problem the tool addresses: a model or tool that is free to download, or free to fork, does not necessarily meet the Open Source Initiative's Open Source Definition. The Gemini CLI case is a textbook illustration. Its Apache 2.0 license scores high on the code-availability dimension. Its backend dependency on Google infrastructure — the authentication, the API, the quota system — would score the overall system lower on the component that matters most to developers who want to contribute with some assurance of continued access: the ability to actually run what they helped build.
LWN.net's May 28 weekly edition featured the MOT alongside the Linux kernel community's parallel discussion about which LLMs can be trusted for security-sensitive code review. Konstantin Ryabitsev of the Linux Foundation raised the same concern in a different context: depending on proprietary large language models for kernel patch review means trusting infrastructure that may change, close access, or alter behavior without notice. The common thread is the same question Alberti asked on GitHub — and the Linux Foundation's tool is one of the few instruments that attempts to answer it before the damage is done.
A Pattern 300 Discontinued Products Long
The Gemini CLI episode joins a list of Google closures that has shaped developer trust for years. The website Killed by Google catalogs more than 300 discontinued Google products and services, a tally that has become cultural shorthand for the risk of building on any Google platform. The pattern is established enough that TechCrunch published an analysis of Stadia's 2022 shutdown under the headline "Stadia died because no one trusts Google."
For developers who contribute to corporate-backed open-source projects, this case adds a dimension that simple product cancellations do not: here, the community's contributions were accepted, the product was improved by volunteer labor, and then the product was handed to paying customers while the contributors lost access. The tool did not fail. It succeeded — and the success was captured by enterprise.
What Developers Must Do Before June 18
Developers relying on Gemini CLI on any non-enterprise tier have until June 18 to migrate. Enterprise integrations at Dynatrace, Elastic, Figma, Shopify, Stripe, and dozens of other launch partners face binary-name changes and configuration rewrites with no extended runway.
Those unwilling to accept a proprietary replacement are evaluating Claude Code, GitHub Copilot CLI integrations, and locally hosted open-source alternatives. Before investing in any corporate-backed open-source AI tool going forward, the Linux Foundation's Model Openness Tool offers a structured way to assess what is actually guaranteed — and what is not — before contributing the first commit. A permissive license on a repository whose backend is controlled by a vendor is not open source in any meaningful sense. The question is not who owns the code. It is who controls the infrastructure the code depends on.
Frequently Asked Questions
Why is Google shutting down Gemini CLI for free users?
Google announced on May 19, 2026 that it is transitioning its developer tooling to a single platform called Google Antigravity. Starting June 18, 2026, free users, Google AI Pro and Ultra subscribers, and individual Gemini Code Assist users will lose API access to Gemini CLI. Enterprise customers with paid Gemini Code Assist Standard or Enterprise licenses are not affected and retain access alongside the new Antigravity CLI.
Why are developers calling the Gemini CLI closure a bait-and-switch?
Google accepted more than 6,000 merged pull requests from external community contributors over nearly a year — explicitly citing those contributions in its own announcement as evidence of success — and then restricted the tool's future to paying enterprise customers while offering free users only a closed-source replacement they had no part in building. Contributor Andrea Alberti asked in the GitHub discussion thread whether contributors had been "essentially working for free on a code base that will only be used in enterprises."
What is the Linux Foundation Model Openness Tool?
The Model Openness Tool, available at isitopen.ai, scores any AI model or tool across 17 components — including model weights, training data, fine-tuning scripts, inference code, and licensing terms — and assigns a Class 1 (closed) through Class 3 (fully open) score. It was developed by researchers at the Linux Foundation, the University of Oxford, Columbia University, and IBM specifically to expose the gap between open-source marketing claims and actual openness. The Gemini CLI case — code available under Apache 2.0, but non-functional without a proprietary Google backend — illustrates exactly the kind of partial openness the tool is designed to surface.
What should I migrate to before the June 18 deadline?
Google's designated replacement is Antigravity CLI, available now but without full feature parity and with tighter weekly usage quotas that users have reported exhausting quickly. Open-source alternatives include Claude Code, GitHub Copilot CLI integrations, and locally hosted models via tools such as Ollama. Before committing to any corporate-backed open-source AI project in the future, the Linux Foundation's Model Openness Tool at isitopen.ai provides a structured way to assess what a project's openness actually guarantees versus what depends on vendor-controlled infrastructure.