Claude Opus 5.5 and Claude Sonnet 5.5 are now available in Google

Claude Opus 5.5 and Claude Sonnet 5.5 are now available in Google Antigravity for Google Cloud customers, according to Google Cloud Blog's report published on October 21, 2026. Marialaura Garcia, Product Marketing Manager, Google Cloud, and Rich Hyndman, Developer Relations, Google Antigravity, write that a quick bug fix and a complex migration call for different models. Claude Opus 5.5 is built for complex work. Claude Sonnet 5.5 is a faster, lower-cost complement. Once an administrator enables them, developers select either Claude model from the model picker, and administrators can also select Gemini 3.8 Flash and Gemini 4 Argon.

This piece is for Antigravity developers and for administrators on Gemini for business or Gemini platform who authorize partner models. It covers the rollout on Antigravity 2.0, the Antigravity CLI, and the IDE extensions, then two patterns in one agent harness: LLM-as-a-Judge verification, and Flash subagents that set inherit, flash, or pro in .agents/agents files. It also covers pay-as-you-go billing on the existing Google Cloud Billing account and the first-party Gemini tokens that still run in the background for coordination.

What shipped in Claude Opus 5.5 and Claude Sonnet 5.5 are now

Claude Opus 5.5 and Claude Sonnet 5.5 are available in Google Antigravity for Google Cloud customers. Claude Opus 5.5 is for complex work, including a multi-file change designed in planning mode. Claude Sonnet 5.5 is the faster, lower-cost complement. Gemini 4 Argon is the other planning-mode choice for that Model A step. Gemini 3.8 Flash is among the Gemini models administrators can select. The Claude names show in the model picker only after an administrator enables partner models for the organization.

The clients are Antigravity 2.0, the Antigravity CLI, and the recently released Antigravity IDE extensions for VS Code, JetBrains IntelliJ, and Visual Studio. Google Cloud keeps the model choice inside a unified enterprise billing and security boundary. A developer can switch the active model between turns in the same conversation thread. Workspace state, open artifacts, and terminal context stay in place. Lightweight first-party Gemini models still run in the background for high-frequency client-side coordination, including context management and summaries. They consume first-party tokens and can incur charges.

What improved in Claude Opus 5.5 and Claude Sonnet 5.5 are now

Pattern 1 is cross-model verification, labeled LLM-as-a-Judge, for agentic loops and multi-step graph workflows. Same-model review can reinforce blind spots. An implementer paired with another family "provides independent verification before moving to the next stage of a workflow." Teams run that check interactively or through automated subagents. Model A plans on Claude Opus 5.5 or Gemini 4 Argon and implements a multi-file change. Before a commit or the next loop step, Model B checks the work against the test suite and the requirements, via a model switch or a one-shot /model check in the CLI.

Pattern 2 pairs a few high-reasoning decisions with dozens of high-volume mechanical lookups. The report lists searching files, grepping logs, checking lint output, and classifying test failures. "Running every lookup on a frontier model can increase both latency and spend." Files under .agents/agents/*.md set inherit, flash, or pro while the primary session stays on the more powerful model. Inherit uses the active session model for deep reasoning. Flash uses Gemini Flash for search, classification, and triage. The sample pro agent audits with Gemini Pro.

What you gain from Claude Opus 5.5 and Claude Sonnet 5.5 are now

The fast loop is .agents/agents/test-triager.md, under the name test-triager. Its description says it runs unit tests, classifies failures, and returns a concise summary of failing assertions. The model field is flash, so the burst uses Gemini Flash rather than the session model. Tools are read and edit. The agent runs the requested test targets, classifies each failure as a compilation error, an assertion failure, or a timeout, and returns a structured summary to the parent agent. The primary session keeps the reasoning model.

The audit file is .agents/agents/gemini-judge.md, and the name field is cross-model-judge. It is an independent LLM-as-a-judge verifier that audits diffs and edge cases using Gemini Pro, with model set to pro and the only tool set to read. It checks the current workspace diff against the user's requirements for regressions, missing edge-case tests, and security issues, then outputs a pass/fail verdict with specific line references. test-triager runs the test loop, and cross-model-judge validates the final diff as steps in one workflow.

How to get Claude Opus 5.5 and Claude Sonnet 5.5 are now

Administrators enable partner models before developers select them. On Gemini for business, the console path is Gemini Enterprise, then Settings, then AI developer tools, for overage billing, spend limits, and authorization. The named guide is Configure AI developer tools settings. Usage is pay-as-you-go on the existing Google Cloud Billing account, outside the pooled Gemini Enterprise Standard and Plus edition monthly token quota, and overages must be enabled. On Gemini platform, with Antigravity through the API, enable Claude models in Model Garden and complete the standard Anthropic terms.

Developers update to the latest Antigravity 2.0, the Antigravity CLI, or the IDE extensions for VS Code, JetBrains IntelliJ, and Visual Studio. They open the model picker or run /model and select Claude Opus 5.5 or Claude Sonnet 5.5. A one-shot /model check lets another family critique the change before a commit. Switching between turns keeps workspace state, open artifacts, and terminal context. The files below pin the subagent: flash for test-triager, pro for cross-model-judge, and inherit to follow the active session model.

/model

---
name: test-triager
description: Runs unit tests, classifies failures, and returns a concise summary of failing assertions.
model: flash
tools:
- read
- edit
---
Run the requested test targets, classify any failures by root cause (compilation error, assertion failure, or timeout), and return a structured summary to the parent agent.

---
name:cross-model-judge
description: Independent LLM-as-a-judge verifier that audits diffs and edge cases using Gemini Pro.
model: pro
tools:
- read
---
Review the current workspace diff against the user's requirements. Check for regressions, missing edge-case tests, and security issues. Output a pass/fail verdict with specific line references.

What to watch after Claude Opus 5.5 and Claude Sonnet 5.5 are now

The models are available now for Google Cloud customers on Antigravity after an administrator enables partner models. Claude Opus 5.5 and Claude Sonnet 5.5 bill pay-as-you-go on the Google Cloud Billing account, outside the pooled Gemini Enterprise Standard and Plus edition monthly token quota, with overages enabled and a spend limit under AI developer tools. Background first-party Gemini models still do context management and summaries, consume first-party tokens, and can still incur charges while the picker shows a Claude model.

The judge has to be a different model family from the implementer. Claude Opus 5.5 or Gemini 4 Argon can plan the change, and the other family critiques it through a model switch or a one-shot /model check against the test suite. A review that stays on the writing model can reinforce blind spots. test-triager may read and edit. cross-model-judge is read-only and returns a pass/fail verdict with line references. Flash takes the mechanical lookups. Inherit keeps deep reasoning on the active session model.