On 18 June 2026, Google shut down the free tier of Gemini CLI — the open-source terminal AI agent it had launched to compete with Claude Code and OpenCode — and replaced it with something called Antigravity CLI. The replacement is a closed-source Go rewrite. It doesn’t have feature parity at launch. The community’s response, as documented by the Linux Foundation and covered across developer forums, has been blunt: a “sickening bait-and-switch.”
What happened
Gemini CLI launched as an open-source project with a generous free tier: 60 model requests per minute, 1,000 per day, access to Gemini 2.5 Pro’s one-million-token context window. It accumulated a significant developer following quickly — not least because the free allowance made it genuinely usable for real work without a subscription.
The June 18 shutdown affects free, Pro, and Ultra users alike. Continued access requires a paid Gemini Enterprise Agent Platform API key. The replacement — Antigravity CLI — is closed-source, not yet at feature parity, and carries none of the naming or community continuity of the product teams adopted.
The platform risk lesson
This episode is a clean example of a pattern worth understanding before committing workflows to any AI tool: free tiers and open-source positioning are often subsidies to acquire users. They can be changed unilaterally once adoption is established.
That’s not a reason to avoid all commercial AI tools — most of the best development tooling in 2026 is commercial. But it’s a reason to think about how embedded a tool becomes before the terms shift. A development team that’s spent months building workflows around Gemini CLI now has to absorb either migration costs or a new payment obligation.
Questions worth asking before committing to an AI coding tool
- What happens to our workflow if the pricing model changes tomorrow?
- Is the tool’s core value proposition stable regardless of monetisation shifts?
- How portable is the configuration and workflow if we need to switch?
The teams least affected by the Gemini CLI shutdown were those using provider-agnostic harnesses — tools like OpenCode that route to multiple model providers and don’t build workflow lock-in around a single vendor’s CLI. When one provider changes terms, swapping the model out is a config change rather than a migration project.
Platform risk in AI tooling is real and 2026 is still early enough that the landscape is shifting fast. The Antigravity episode is a useful reminder to build in resilience before you need it.
We cover which AI development tools we use in production on our AI-assisted development page.