Migrating to Claude Code as an Alternative to Antigravity’s Pay-As-You-Go Mode on GCP¶
I found out through a pop-up in Antigravity’s own interface that the tool would stop being included in corporate Google Workspace licenses. The change took effect on July 7, 2026, and my work account was capped at the same quota as any free user. I had already written enthusiastically about what’s new in Antigravity 2, but this billing change made me rethink my entire agentic development stack, both at work and personally.
The cutback wasn’t a minor quota adjustment. As documented in a Google AI community thread, Google notified Workspace administrators, around May 5, 2026, that it would discontinue the AI Ultra Access add-on — the Workspace plan (~\(250/month) that included the Antigravity, Gemini CLI, and Gemini Code Assist plugin developer tools. Without administrator action, the organization automatically migrates to the much cheaper AI Expanded Access add-on (\)24/month), which doesn’t include those developer tools. The change applies from July 7, 2026 for flexible plans, and from the next renewal on or after that date for annual plans. Losing that add-on leaves no middle ground: there is no paid corporate subscription plan for Antigravity as such.
Pay-as-you-go: higher bills, no Claude¶
The change forces a choice between two paths, and neither is free or convenient. The first is enabling consumption-based billing: activate the Agent Platform API on your own or your organization’s Google Cloud (GCP) project, link a billing account, and sign in to Antigravity using that project as the backing account. Usage then gets billed per token and per compute time in the agent’s sandbox, instead of being bundled with the Workspace license — the bill goes up, no exceptions.
But the real problem with this path isn’t just the added cost: per Antigravity’s model documentation, Enterprise plans — the ones consumption mode falls under — only offer Gemini models. Anthropic models are available exclusively on personal plans. For any corporate team relying on Claude for its agentic workflow, this is the critical point — the “official” path to keep using Antigravity as a business leaves you, in practice, without access to Anthropic.
It’s a real SLA problem: Google acts as a router to a third party’s infrastructure that it doesn’t host or operate directly, and that’s why it can’t extend to its Enterprise customers the same service guarantee it would offer over its own infrastructure.
Personal accounts: Claude 4.6, no update in sight¶
The second path is skipping consumption mode and sticking with a paid personal account. Claude does show up in the picker there — but the newest available Anthropic model is still Claude 4.6, with no mention of an upcoming update, while Anthropic is already three versions ahead: 4.7, 4.8, and 5.
This is the most frustrating part, because Antigravity’s developer experience still feels better than Claude Code’s in several ways: native multirepo project support, per-conversation worktree management with no manual intervention, the ability to leave inline comments directly on the agent’s artifacts, and standards already shared with Claude Code like AGENTS.md. But all of those tooling advantages become irrelevant — even counterproductive — if the model behind them isn’t up to the task. A brilliant IDE orchestrating an outdated model still produces worse code than a more austere CLI orchestrating the best model available.
Given the lack of up-to-date Anthropic models — whether because consumption mode simply doesn’t offer Claude, or because the personal account leaves it stuck on 4.6 — I seriously tried using Gemini Flash and Gemini Pro as the main engine inside Antigravity for real development tasks. The experience wasn’t pleasant: on refactoring tasks with broad context, on following project-specific code style instructions, and on the overall quality of proposed diffs, both models consistently landed below what I was already getting from Claude in the same IDE. This isn’t a blanket dismissal of Google’s models — they likely perform well for other tasks — but for the kind of agentic coding work I do daily, the difference was noticeable enough to rule them out as a primary option.
My decision: migrating to Claude Code¶
With both paths exhausted — consumption mode without Claude, or a personal account with Claude stuck in place — the decision was simple, if uncomfortable: I’m migrating my workflow, both personal and corporate, to Claude Code.
I’d rather deal with the learning curve and friction of stricter step-by-step control, backed by an up-to-date model, than enjoy a superior IDE tied to a model catalog that either doesn’t include Claude at all or leaves it outdated. This doesn’t close the door on Antigravity — if Google wires Claude into consumption mode and updates the personal-account cap, the equation could change again. But for now, the relationship between tooling and model is what decides which tool I use, not the other way around.
Conclusion¶
Antigravity’s cutback for Workspace accounts on July 7 was just the entry point to a deeper problem: neither consumption mode nor a personal account gets you an up-to-date Claude inside Antigravity. That limitation is an SLA problem — a consequence of depending on a third party Google doesn’t operate directly — and a matter of Google’s product roadmap. As long as that gap exists, no amount of tooling advantage — however good — makes up for working without the best model available.
References¶
Antigravity seems to be soft-deprecated. Google AI Developers Forum.
Antigravity 2: What’s New for Agent-Driven Development. Cosmoscalibur.