Google has fundamentally reimagined how developers write code by replacing the traditional editor-centric model with a standalone application built entirely around orchestrating multiple AI agents working in parallel. Announced on May 19, 2026, at Google I/O, Antigravity 2.0 marks a generational shift in the AI coding tool landscape, positioning autonomous agents as first-class teammates rather than sidebar assistants. The platform ships with a new command-line interface, a software development kit, native Android support, and a pricing structure that includes a new $100-per-month AI Ultra plan and a price reduction on the top tier from $250 to $200.

What Changed Between Antigravity 1.0 and 2.0?

The original Antigravity, launched in November 2025, wrapped AI agents around a conventional code editor in a familiar sidebar-assistant model. Version 2.0 inverts that relationship entirely. The new desktop application is built "entirely around agent orchestration," according to Google's framing, treating the editor as one tool among several rather than the center of the development experience. This architectural shift changes the fundamental unit of work: instead of editing a file line by line while an AI model suggests the next few lines, developers now describe an outcome and dispatch one or more agents to achieve it, with those agents writing code, launching applications, driving a browser, verifying results, and reporting back with tangible deliverables.

The competitive implication is significant. Cursor, Windsurf, and GitHub Copilot all began as editor-first experiences that grew agent features. By declaring the standalone agent app the primary product, Google is betting that the next abstraction layer is orchestration, running and supervising many agents at once, and that whoever owns that layer owns the developer relationship.

How Does Gemini 3.5 Flash Power the Speed Advantage?

Underpinning Antigravity 2.0 is Gemini 3.5 Flash, set as the default model across the platform. Google states that the model was co-developed using Antigravity itself, positioning the tool as production-grade enough to help build the very model that powers it. On performance, Google claims that Gemini 3.5 Flash "outperforms Gemini 3.1 Pro across almost all benchmarks while running four times faster than other frontier models". While these are vendor benchmarks and should be read with appropriate caution, the strategic choice matters: speed is not a vanity metric in agentic development. When a single feature request spawns a plan, several parallel subagents, browser verification, and a test run, latency compounds at every step. A model that is materially faster per token changes the economics of running dozens of agents concurrently, which is precisely the workflow Antigravity 2.0 is designed around.

Google

Choosing a "Flash" tier model as the default, rather than the flagship "Pro," signals that Google is optimizing for throughput and cost across many agents rather than maximum reasoning on a single thread. This also decoupled Google's developer-tools story from the flagship Gemini 3.5 Pro model's timeline, which faced a widely reported schedule slip into July, allowing Google to own the I/O 2026 developer narrative regardless of Pro's availability.

How Does Antigravity 2.0's Pricing Compare to Competitors?

Pricing is where Antigravity 2.0 becomes strategically interesting for teams trying to budget. Individuals can still use Antigravity for free, as they could at the November 2025 launch. Google restructured its paid Google AI subscriptions to grant higher Antigravity usage limits, introducing a new $100-per-month AI Ultra plan offering five times the usage limits of the existing Google AI Pro plan. Most notably, Google cut the price of its top AI Ultra tier from $250 to $200, which grants twenty times the Pro plan's limits. That price cut runs against the grain of a market that has spent 2026 raising prices: GitHub Copilot moved to usage-based billing on June 1, making Google's move unusual enough to read as a deliberate land-grab.

This aggressive pricing strategy reflects Google's confidence in the efficiency of Gemini 3.5 Flash and the scale of its infrastructure. The company is using these advantages to undercut rivals on the high end while capturing free-tier individuals at the bottom, a two-pronged approach designed to lock in market share during a critical moment when the industry is consolidating around agent-first development.

What Are the Key Features and Capabilities of Antigravity 2.0?

• Agent Orchestration: Multiple AI agents can plan, write, run, and verify code in parallel, with developers able to spawn agents, run tasks concurrently, schedule background work, and design custom subagent workflows.
• Multi-Platform Support: Available on macOS, Windows, and Linux, with native Android app builds, extending the platform beyond traditional desktop development environments.
• Voice Commands and Integration: Native voice commands, an Antigravity export tool inside AI Studio, and tighter integration with Firebase and Android reinforce that the platform is now a hub for shipping production software rather than a smarter autocomplete.
• Model Flexibility: Supports Gemini 3.5 Flash as the default, along with Gemini 3.1 Pro, Anthropic Claude Sonnet 4.5, and OpenAI GPT-OSS, giving developers choice in which models power their agents.
• Enterprise Offering: Gemini Enterprise Agent Platform on Google Cloud provides dedicated infrastructure and compliance features for large organizations.

How Can Developers Get Started With Antigravity 2.0?

For developers looking to integrate Antigravity with existing tools, the integration process is straightforward. Composio, a tool integration platform, offers a guide for connecting Ollama (a local language model runner) to Antigravity via Composio Connect. The process involves opening Antigravity Settings, navigating to the Customizations tab, clicking "Open MCP Config" to access the mcp_config.json file, and adding the Composio server configuration with an API key. After saving the file and refreshing the Installed MCP Servers section, developers can ask the Antigravity agent to connect to Ollama or perform Ollama-related tasks such as running custom prompts with llama2, getting model information for all installed models, or generating text completions using mistral. This approach allows developers to leverage local or cloud-based language models without exposing credentials in prompts or local scripts.

The broader context matters here: Antigravity 2.0 arrives at a moment when the AI coding tool market is fragmenting along both technical and geopolitical lines. While Google positions itself as the agent-orchestration leader, competitors like Z.ai are launching their own agent-first environments powered by open-source models trained entirely on non-American chips, reflecting growing concerns about vendor lock-in and sovereign access risk. The market is moving fast, and the winner will likely be whoever can convince developers that their abstraction layer, their model, and their pricing model best serve the future of software development.

"Antigravity 2.0 is built entirely around agent orchestration," Google stated in its launch materials, emphasizing that agents deserve a dedicated space to work rather than living as a chatbot in a sidebar.

Google, Antigravity 2.0 Launch Documentation

For Irish developers and teams watching the agentic coding race from Dublin's dense cluster of multinational engineering offices, Google Antigravity is no longer a curiosity bolted onto a code editor. It is Google's clearest statement yet that the integrated development environment itself is being demoted, and that the primary abstraction for building software is shifting from files-and-cursor to fleets of orchestrated agents. The question now is not whether agents will become central to development, but which platform will own that relationship.