Claude 5.5 in Google Antigravity: Model choice changes the bill

Why Google is opening its agent environment to Claude

Antigravity is Google's agentic coding environment, long associated with Google's Gemini models. Eligible Google Cloud customers can now select Anthropic's competing Claude Opus 5.5 and Sonnet 5.5 alongside Gemini. Google positions Opus as capable of more complex tasks and Sonnet as a faster, lower-cost complement. These are product descriptions, not a neutral head-to-head result. The underlying shift is strategic: a coding environment is becoming a platform that manages models from multiple vendors while keeping the developer's workspace and workflow intact. Each model may still carry different obligations and operational constraints.

Model switching without starting over

The new selection applies to Antigravity 2.0, its command-line interface and IDE integrations such as VS Code and JetBrains. Google's announcement says developers can switch the active model between turns while preserving open artifacts, terminal state and workspace context. An engineering team could therefore choose one model for routine tests and another for a difficult migration. But seamless access is not the same as independently measured productivity. Teams need to inspect generated code, account for retries and compare the actual quality of accepted changes rather than assuming that two models are always better than one.

The billing rule companies must not overlook

According to Google's administrator guidance, partner models do not use the pooled Gemini Enterprise Standard and Plus monthly token allowance. Instead, usage is charged to the existing Google Cloud Billing account on a pay-as-you-go basis. Organizations may need to authorize overages, configure spending caps and accept partner terms. That changes the economics of experimenting with model choices. Token pricing alone is not a useful enterprise measure unless it also captures background coordination, tool calls, failed attempts and developer time spent correcting the output.

A shared interface does not erase governance differences

Google Cloud documentation says third-party models are disabled by default until an administrator opts in. Access can later be revoked. Release notes reference global, US and EU multi-region inference availability, which is not the same as a universal assurance of in-country processing throughout Europe. Enterprises with regulated data must check each model's terms, access controls, auditability and geographic commitments. The platform may provide a common entry point, but it does not make the providers' compliance responsibilities interchangeable. Context sharing and credentials deserve careful review whenever a workflow spans model vendors.

The practical experiment worth running

Teams should compare Claude and Gemini against repeatable jobs such as migrations, test generation and review. Track the cost per accepted change, failure rate, latency, quality and amount of human correction. Set model-specific permissions for sensitive repositories and audit spending at the workflow level. AgentenCode sees the broader story as the rise of the multi-model coding agent, not a contest with a guaranteed winner. The benefit of optionality is real only if companies preserve control of data, evaluation and costs while moving between models.

Original sources

• Google Cloud: Claude in Antigravity ↗
• Google Cloud: Gemini Enterprise release notes ↗
• Google Cloud: AI developer tool settings ↗