Google Antigravity 2.0: Free AI Coding Agents That Work in Parallel

Google just launched a free coding tool that runs multiple AI agents simultaneously. It supports Gemini, Claude, and GPT models. Here is how it compares to Cursor, Claude Code, and Codex.

• For the last two years, serious AI development on laptops was dominated by two tools: Claude Code from Anthropic and Codex from OpenAI. Google just changed that conversation.
• At Google I/O 2026 on May 19, Google launched Antigravity 2.0, a standalone desktop application that runs multiple AI agents in parallel to plan, write, debug, and ship code. Unlike its predecessor which looked like another code editor with an AI sidebar, version 2.0 centers the agent instead of the editor. It is free to start, runs on Gemini 3.5 Flash (which Google claims is 4x faster than competing frontier models), and supports Claude and GPT models alongside Gemini.
• Here is everything you need to know: what it does, how it compares, what it costs, and whether you should switch.

What Makes Antigravity 2.0 Different

The core difference from every other AI coding tool is the multi agent architecture. When you give Antigravity a task, it does not just send one prompt to one model. It spins up a team of specialized sub agents that work simultaneously.

At the I/O keynote, Google demonstrated this by using 93 sub agents working in parallel for 12 hours to build an entire working operating system from scratch. In more practical demos, users prompted it to build a D2C brand website. Multiple agents fired up in parallel: a Markup Architect handling HTML structure, a Style Stylist generating brand images, and a debugging agent catching errors in real time.

Antigravity 2.0 has two views. The Editor view works like a traditional IDE with an agent sidebar, similar to Cursor or Windsurf. The Manager view is new: a clean chat interface where you describe what you want in plain English and agents handle the rest. You use commands like /go for management tasks and /browser to see the live working website.

The platform also includes a Schedule Task feature that lets agents run automated daily tasks like website audits or competitor research, essentially turning the tool into an autonomous employee.

For developers who prefer their own code editor, there is an Antigravity CLI (invoked as agy in the terminal) that installs into VS Code, JetBrains, or any terminal. It supports spawning specialized sub agents like a code integrity auditor or mobile web tester.

Source: Verified from Google I/O 2026 keynote, Google Developer Blog, TechCrunch, and The Next Web, May 2026.

How It Compares to Cursor, Claude Code, and Codex

The AI coding tool market now has five serious contenders. Here is how they stack up.

Antigravity 2.0 is the only tool that natively runs multiple agents in parallel. Claude Code supports agents but requires manual setup. Cursor uses a single model per request. Codex is locked behind a paid OpenAI subscription. Kilo Code supports BYOK with 500+ models but does not have the multi agent orchestration.

On speed, Google claims Gemini 3.5 Flash generates output at up to 800 tokens per second, making it roughly 4x to 12x faster than competing frontier models depending on the task. This speed advantage is real for iteration heavy workflows where you are constantly generating, reviewing, and regenerating code.

On code quality, Claude Code with Opus 4.7 still leads on complex reasoning tasks. The SWE Bench Pro benchmark puts Claude at 64.3%, ahead of Gemini 3.5 Flash. For straightforward code generation and scaffolding, Antigravity is fast and capable. For production grade refactoring of complex codebases, Claude Code remains the stronger option.

On model flexibility, Antigravity supports Gemini 3.5 Flash, Gemini 3.1 Pro, Claude Sonnet 4.6, Claude Opus 4.6, and GPT OSS 120B in the free tier. Kilo Code also supports BYOK with 500+ models. Cursor and Codex are more locked into their respective ecosystems.

Pricing: Free Tier With Serious Limits

Antigravity 2.0 has a free tier, but there is important context about how Google has handled free access historically.

When the original Antigravity launched in November 2025, the free tier offered 250 requests per day. By December 2025, that was cut to 20 requests per day, a 92 percent reduction. In March 2026, Google replaced request limits with a credit based system. The free tier now uses compute based usage that refreshes every 5 hours.

The paid tiers are competitive with the rest of the market. AI Pro costs $19.99 per month (same as Google One AI Premium). Google introduced a new AI Ultra tier at $100 per month with 5x higher limits, and reduced the top tier from $249.99 to $200 per month with 20x limits. Credits cost $0.01 each and can be purchased in bulk ($199 for 20,000) to continue working when your quota runs out.

For comparison, Cursor costs $20 per month, Claude Code Pro is $20 per month, and Codex is included in ChatGPT Plus at $20 per month. The pricing is converging across the industry at approximately $20 per month for the entry tier and $100 to $200 for power users.

Should You Switch

Here is the practical advice based on your situation.

If you are new to AI coding tools: Start with Antigravity 2.0's free tier. It is the lowest barrier to entry and the multi agent workflow is genuinely impressive for scaffolding new projects. Download it from antigravity.google for Mac, Windows, or Linux.

If you use Cursor or Copilot: Try Antigravity alongside your current tool, do not switch completely. The multi agent parallel execution is unique and worth testing on your actual codebase. Keep your existing tool as a fallback since Google's free tier limits have been unreliable historically.

If you use Claude Code: Keep it. Claude Opus 4.7 still produces the best code quality on complex tasks. Use Antigravity for rapid prototyping and scaffolding where speed matters more than precision. The two tools complement each other.

If budget is your primary concern: Antigravity's free tier and Kilo Code's open source extension are the best options. Kilo gives you BYOK model flexibility. Antigravity gives you multi agent orchestration.

One important caveat: Google has confirmed that consumer access to Gemini CLI and Gemini Code Assist IDE extensions will end on June 18, 2026. Antigravity is now the only path forward for Google AI coding tools. If you are currently using Gemini CLI, migrate to Antigravity CLI before that date.

Not sure which coding tool fits your workflow? Take our free 60 second AI Match quiz: aitoolsmentor.com/wizard

Source: Pricing verified from antigravity.google, cursor.com, claude.ai/pricing, and chatgpt.com/pricing, May 2026.