Google Antigravity Unleashed: Now with Claude 4.5 Opus and Higher Rate Limits for All

Google’s ambitious AI-powered IDE, Google Antigravity, has just received a massive upgrade, addressing some of the most significant user concerns and adding powerful new capabilities. Previously, while impressive, the tool was hampered by restrictive rate limits that applied even to paying Google subscribers. Now, Google has revamped its limits for both free and paid users and introduced support for one of the most advanced AI models available today: Claude 4.5 Opus. Let’s dive into what’s new and how you can leverage these changes to supercharge your coding workflow.

[00:08.529]

When Google Antigravity first launched, it presented itself as a next-generation, agentic development platform designed to evolve the IDE into an “agent-first era.” It offered an impressive suite of features, including an AI-powered editor with tab autocompletion, natural language code commands, and a configurable, context-aware agent. However, its practical use was often cut short by strict rate limits, which were imposed to keep the service free. This meant that even if you had a premium Google AI subscription, you saw no benefit when using Antigravity, which was a major frustration for many developers.

[00:41.879]

In a welcome update, Google has officially announced higher and more generous rate limits for Google Antigravity. This resolves one of the biggest pain points for developers. The update brings significant improvements for all users, whether they are on a free or paid plan.

Google AI Pro and Ultra subscribers now receive priority access, featuring our highest, most generous rate limits with quotas that refresh every five hours. For users on the free plan, we’ve shifted to a larger, weekly based rate limit to minimize hitting rate limits quickly during a project.

This is a game-changer. Paid subscribers now get a substantial boost with quotas refreshing every five hours, making extended coding sessions much smoother. Free users also benefit from a new weekly-based limit, which should provide more flexibility and prevent abrupt interruptions during a project. While the exact numbers aren’t public, the overall experience feels significantly more generous.

[01:37.899]

Perhaps the most exciting new feature is the integration of Claude 4.5 Opus. This state-of-the-art model is now available directly within Antigravity for all users, including those on the free tier. This is a massive value proposition, as Opus is an expensive model to use via its API. Antigravity provides access with very reasonable limits, giving developers the power to combine the strengths of different AI models. You can now use the best tool for the job: Gemini 3 Pro for its excellent frontend capabilities and Claude 4.5 Opus for its superior planning, backend logic, and debugging skills.

[02:58.559]

To get the most out of these new capabilities, a strategic workflow is key. The demo showcases a powerful method using a free account. The process begins by giving the AI a clear, structured prompt to plan a new feature. The recommended approach is to use Claude 4.5 Opus for the initial planning phase due to its exceptional reasoning abilities. A great technique is to ask the model to “bifurcate” the tasks into distinct Backend and Frontend categories, and to prioritize the backend implementation first.

Here is the example prompt used in the video:

Create a settings page and add an option which allows the user to set a custom system prompt for all their messages. To implement this feature please bifurcate it into Backend and Frontend tasks. (Denote Backend tasks as B1, B2 etc. while denote Frontend Tasks as F1, F2). I'd like to implement the backend tasks first and then move to frontend. So, please plan the tasks accordingly.

[04:52.349]

Once the prompt is submitted, Opus generates a detailed implementation plan, neatly organized into Backend Tasks (like adding a mutation in Convex and integrating the system prompt into the chat API) and Frontend Tasks (like creating UI components and a state management store). With this plan in place, you can proceed with implementation. The workflow is simple:

1. Implement Backend: Stick with Claude 4.5 Opus to execute the backend tasks (B1, B2, etc.).
2. Implement Frontend: Switch the model to Gemini 3 Pro to handle the frontend tasks (F1, F2, etc.).
3. Debug: If you encounter any bugs, switch back to Opus for its powerful debugging capabilities.

[05:55.779]

During the implementation process, Antigravity’s agent works autonomously, creating and editing files as needed. If it runs into issues, like the type error shown in the video, it’s smart enough to diagnose the problem and even suggest and run terminal commands to fix it. In this case, it identified that the Convex generated code needed to be regenerated and ran the necessary command.

npx convex dev --once

[06:22.099]

After both the backend and frontend tasks are completed, the AI generates a final walkthrough document. This serves as a comprehensive summary of the changes, including implementation details and manual verification steps. It’s an incredibly useful artifact for reviewing the work and ensuring everything functions as expected. This structured approach, combining automated checks like TypeScript compilation with manual verification guidance, helps build trust in the AI agent’s work.

[06:37.769]

The demo successfully implements a Custom System Prompt feature, which allows users to configure the AI’s personality. By setting a prompt like “Speak like a drunk pirate,” the user can change how the AI responds in all subsequent messages. This demonstrates the end-to-end capability of Antigravity, from initial planning to a fully functional feature. Furthermore, the video highlights another recent addition: Nano Banana Pro support, which allows the agent to autonomously generate, save, and reference asset images like icons or banners within your project, further streamlining the development of cohesive applications. Overall, these updates make Google Antigravity an incredibly powerful and accessible tool for developers, offering top-tier AI models and generous limits, all for free.