Google is expanding its Antigravity development platform with two new AI models from Anthropic, giving developers more flexibility to build, test, and review code using multiple AI systems within the same workflow.

Announced on October 20, Claude Opus 5.5 and Claude Sonnet 5.5 are now available in Google Antigravity for Google Cloud customers. Google positions Opus as the option for complex engineering tasks, while Sonnet offers a faster, lower-cost alternative for work that does not always require the most advanced reasoning capabilities.

The addition gives developers another choice alongside Google’s Gemini models, reflecting a broader shift toward development environments where multiple AI models collaborate rather than relying on a single model for every task.

Claude Opus 5.5 and Sonnet 5.5 come to Antigravity

Developers can select the two Claude models from Antigravity’s model picker once an administrator enables access. The models are supported across Antigravity 2.0, the Antigravity command-line interface (CLI), and its IDE extensions for Visual Studio Code, JetBrains IntelliJ, and Visual Studio.

Google says the integration allows teams to choose models based on the demands of individual tasks while keeping enterprise billing and security controls within their existing environment.

That flexibility can be particularly useful for software projects involving several stages of development. A complex change might require one model to plan an implementation, another to inspect the resulting code, and a faster model to handle repetitive tasks such as searching files or classifying test failures.

Rather than treating model selection as an all-or-nothing decision, Antigravity can bring different models into the same development process.

How multi-agent workflows can benefit from multiple models

Google highlights two engineering patterns that demonstrate how developers can use different models together in Antigravity.

The first is cross-model verification, sometimes described as an LLM-as-a-Judge workflow. Instead of asking the same model to write code and then review its own work, developers can use one model to implement a change and another to evaluate the result against requirements and tests.

For example, a developer could use Claude Opus 5.5 or Google’s Gemini 4 Argon to plan and implement a change involving multiple files. A separate model could then examine the implementation, identify potential regressions, and check whether important edge cases have been addressed before the developer proceeds.

Using different model families for implementation and review can provide an additional perspective, although it does not guarantee that the resulting code is correct or free of security issues.

The second pattern combines reasoning models with faster subagents. Not every step in a software engineering task requires extensive reasoning. Searching a repository, scanning logs, checking lint output, and grouping test failures are often more mechanical than designing an architecture or debugging a complicated problem.

Antigravity’s subagent customization system lets developers assign model tiers such as inherit, flash, and pro to specialized agents. A subagent using inherit follows the active session’s model, while one configured with flash can delegate high-volume searches and classification tasks to Gemini Flash models.

This approach allows developers to reserve more capable reasoning models for demanding decisions while using faster models for routine work. The goal is to balance code quality, response time, and AI usage costs.

What developers need to know about availability

Claude Opus 5.5 and Claude Sonnet 5.5 are available to Google Cloud customers using Antigravity, subject to the relevant account and administrator settings.

For organizations using Gemini Enterprise, administrators must authorize the third-party models and configure pay-as-you-go billing through their existing Google Cloud Billing account. Usage of these partner models does not come out of the pooled monthly token allowance for Gemini Enterprise Standard and Plus editions, and overage billing must be enabled.

Organizations accessing Antigravity through the Gemini platform API can also enable the Claude models in Model Garden and complete Anthropic’s required terms.

Developers can then update to the latest Antigravity 2.0, CLI, or supported IDE extension and select the desired model from the model picker. In the CLI, they can use the /model command. Google also allows developers to switch models between turns while retaining the existing conversation’s workspace state, open artifacts, and terminal context.

One important detail is that Gemini models continue to operate in the background for certain client-side coordination tasks, including context management and summaries. Google notes that these first-party models consume Gemini tokens and may incur additional charges.

By bringing Anthropic’s latest Claude models into its development environment, Google is making Antigravity more flexible for teams that want to combine different AI systems in a single engineering workflow. For developers building multi-agent applications, the benefit is not simply having more models to choose from, but being able to assign each model a role that suits its strengths and the cost of the task.