Gemini 4 Argon in Antigravity: Developer Setup Guide

Google released Gemini 4 Argon on October 13, 2026, as its latest frontier model for complex software engineering, knowledge work, and cybersecurity tasks. For developers, one of the strongest surfaces is Google Antigravity, the agent-first development platform that now supports Argon for Google AI Ultra subscribers.

Antigravity lets Argon run multi-step agentic loops, connect to tools via MCP, and handle long-horizon coding jobs that earlier models could not finish reliably. This guide walks through access requirements, setup, and practical first uses.

If you are building Android apps, you can also call Argon through Firebase AI Logic. See our guide on adding Gemini 4 Argon to Android with Firebase AI Logic.

Access requirements

Gemini 4 Argon in Antigravity requires a Google AI Ultra subscription for individual users. Eligible Gemini Enterprise plans and Google Cloud business plans also grant access. The free individual tier of Antigravity does not include Argon.

Update the Antigravity desktop app or CLI to the latest version. Sign in with the Google account that holds the Ultra subscription. Argon appears as a selectable model for agent tasks once the rollout reaches your account.

Introductory API pricing for Argon runs $2 per million input tokens and $10 per million output tokens through November 30, 2026. After that, rates rise to $4 and $20. Direct API calls bill separately from subscription quotas.

Install and open Antigravity

Download Antigravity from antigravity.google for macOS, Windows, Linux, or Googlebook. The Gemini macOS app also offers an Antigravity coding interface for eligible accounts.

After installation, open the app and sign in. Navigate to the model selector or agent settings and choose Gemini 4 Argon (model ID: gemini-4-argon). Confirm the selection before starting a new agent session.

For CLI users, ensure your environment points to the latest Antigravity binary. Authentication follows the same Google account flow.

Start a simple agent task

Open a project folder in Antigravity. Create a new agent conversation and set the model to Argon. Give it a clear, bounded goal:

“Refactor the authentication module in this repository. Extract the token refresh logic into a separate service, add unit tests for edge cases, and update the README with the new flow. Stop and report when tests pass.”

Argon breaks the work into steps, reads relevant files, proposes edits, and can run terminal commands if you approve them. Monitor the progress panel. Argon shares intermediate status more often than previous models.

Use the /goal command for longer tasks. Provide a one-paragraph objective and let the agent plan and execute overnight if needed. Internal Google teams reported gains in developer velocity when using this pattern with Argon.

Configure thinking levels

Argon supports three thinking levels: low, medium (default), and high. Higher levels improve performance on multi-step reasoning and complex code migrations but increase latency and token use.

In the Antigravity UI, look for the thinking level control in agent or generation settings. For quick classification or extraction, use low. For architecture reviews or cybersecurity analysis, switch to high.

When calling through the Gemini API or Managed Agents, pass thinking_level in the generation config or agent_config. The Interactions API is the recommended surface for new projects.

Use skills and MCP tools

Antigravity supports the Agent Skills format. Place skill folders under your project’s .agents/skills/ directory. Each skill contains a SKILL.md with instructions, scripts, and references.

For global skills stored in ~/.agents/skills/, create a symbolic link so Antigravity discovers them:

ln -s ~/.agents/skills ~/.gemini/config/skills

Restart Antigravity. The skills then appear for Argon to load when relevant.

Connect external tools through the Model Context Protocol (MCP). Examples include Blender for 3D asset generation or custom internal APIs. Argon can loop across dozens of steps, calling tools, inspecting results, and refining output until the goal is met.

Practical tips

Start with medium thinking level and raise it only when results lack depth. Long decode support allows up to 1 million output tokens in a single turn on supported surfaces, useful for generating large codebases or detailed reports.

Break very large tasks into sequential agent runs. Feed the summary of one run as context for the next. This keeps token usage predictable and reduces the chance of context drift.

Review every code change before committing. Argon produces high-quality output but still benefits from human verification on security-sensitive or production paths.

Monitor usage in your Google Cloud billing console. Set alerts so introductory pricing does not become a surprise after November 30.

Conclusion

Gemini 4 Argon is available today in Google Antigravity for Ultra subscribers and eligible enterprise accounts. The combination of frontier reasoning, long output limits, and native agent primitives makes it practical for real software engineering work.

Update Antigravity, select the model, and begin with a focused goal. Pair it with skills and MCP tools for repeatable workflows. For Android-specific integration, the Firebase AI Logic path remains the simplest client-side option.

Sources

• Google Antigravity Blog: Gemini 4 Argon in Google Antigravity (October 13, 2026)
• Google AI Studio: Gemini 4 Argon Developer Guide
• Gemini API documentation for gemini-4-argon and Interactions API