Google Antigravity 2.0: From IDE Plugin to Agent-First Development Platform
At Google I/O on May 19, 2026, Google officially launched Antigravity 2.0 — a standalone desktop application rebuilt entirely around the agent-first philosophy. Alongside it came Antigravity CLI, SDK, and Managed Agents in the Gemini API. This is more than a product upgrade; it is Google's definitive bet on the future of software development.
1. From IDE to Standalone Desktop App: Why Antigravity 2.0?
Antigravity was initially released in November 2025 as a VS Code extension, positioned as an intelligent coding tool directly competing with Cursor. After six months of iteration and user feedback, Google reached a critical conclusion: a true agent-first experience should not be constrained within the boundaries of an IDE.
Antigravity 2.0 is a fully independent desktop application supporting macOS, Linux, and Windows. It no longer depends on any IDE, but exists as an agent orchestration hub. Developers interact with agents synchronously and asynchronously, manage multi-agent parallel tasks, schedule recurring jobs — it's no longer the traditional "open editor, write code, run" workflow.
"We rearchitected the product to be agent-first from the ground up, independent of an IDE or other dev-specific concepts such as repositories."
— Google Antigravity Blog
The original Antigravity IDE and its Agent Manager will be gradually phased out in future releases, with the IDE itself transitioning into a pure intelligence-augmented editor.
2. Antigravity 2.0 Core Features
2.1 Dynamic Subagents
The main agent can dynamically define and invoke sub-agents to complete focused sub-tasks based on task requirements. This design has two key advantages:
- No pollution of the main agent's context window: sub-tasks execute in isolated contexts
- Parallel execution support: multiple sub-agents can run simultaneously, significantly reducing complex task completion time
2.2 Asynchronous Task Management
Tasks and commands can now run asynchronously without blocking the main agent's workflow. You can launch multiple large-scale refactoring tasks simultaneously, let them execute in the background, while the main agent continues processing other requests.
2.3 Scheduled Tasks
You can define cron expressions to trigger agents to execute predefined tasks on schedule. No need to manually start each agent — for example, you can set up automatic code reviews, dependency updates, or test suite runs at midnight every day.
2.4 JSON Hooks
Define hooks in simple JSON format to intercept and control Antigravity agent behavior. This provides a standardized extension point for teams to customize agent workflows and integrate CI/CD processes.
2.5 Projects
Antigravity 2.0 decouples agent conversations from code repositories. Conversations are now organized by "Projects." A project can span multiple folders and have its own agent settings and permissions. This means agents can access broader context for complex tasks while still allowing precise access control.
2.6 New Slash Commands
/goal: Run until a specified task is fully completed without requiring intermediate user input/grill-me: Ask reverse questions to align on specific details before starting implementation/schedule: Execute an instruction as a one-time scheduled task or recurring job/browser: Explicitly control the agent's browser capabilities (based on user feedback, agents sometimes couldn't accurately determine when to use the browser, so it became an explicit command)
2.7 Real-time Voice Transcription
Voice input now uses the latest Gemini Audio model for real-time transcription, instead of passing raw audio files to the model. Speaking is faster than typing — this change directly improves interaction efficiency.
3. Antigravity CLI: Terminal Gateway to Agents
Alongside Antigravity 2.0, Google released Antigravity CLI — a lightweight terminal interface built with Go for invoking, monitoring, and interacting with Antigravity agents.
3.1 Relationship with Antigravity 2.0
| Dimension | Antigravity CLI | Antigravity 2.0 |
|---|---|---|
| Core Positioning | Speed, keyboard efficiency, low overhead | Comprehensiveness, visual orchestration, project management |
| Interface | Terminal User Interface (TUI) | Full rich GUI application |
| Use Case | SSH/remote sessions, keyboard-first | Local workspace, heavy orchestration |
| Agent Engine | Shared core agent engine | Shared core agent engine |
Both share the same Agent Harness, meaning any improvements to the core agent are automatically synchronized to both products. Settings and permissions are also bi-directionally synced.
3.2 Migration from Gemini CLI
Google also announced that Gemini CLI will cease service for free and AI Pro/Ultra users on June 18, 2026. Antigravity CLI preserves Gemini CLI's key features: Agent Skills, Hooks, Subagents, and Extensions (now Antigravity Plugins).
Migration advantages:
- Faster execution: Built with Go, more responsive
- Asynchronous workflows: Multi-agent background orchestration without locking terminal sessions
- Unified architecture: Shares Agent Harness with Antigravity 2.0
4. Gemini 3.5 Flash: A New Model Built for Agents
The core engine powering Antigravity 2.0 is the newly released Gemini 3.5 Flash model, with several key characteristics:
- Outperforms predecessors: Surpasses Gemini 3.1 Pro on nearly all coding, agent tasks, and multimodal reasoning benchmarks
- Blazing fast: Output speed exceeds 280 tokens/second, 70% faster than comparable frontier models
- Co-optimized with Agent Harness: The model and agent framework are co-optimized — a unique advantage for Google
Google DeepMind CTO Koray Kavukcuoglu stated that the model "surpasses our frontier model 3.1 Pro on nearly all benchmarks."
5. Token Pricing and Subscription Plans
5.1 Gemini 3.5 Flash API Pricing
| Type | Price (per 1M tokens) |
|---|---|
| Input (text/image/video/audio) | $1.50 (Global) / $1.65 (Non-global) |
| Output (including reasoning tokens) | $9.00 (Global) / $9.90 (Non-global) |
| Cached input (90% discount) | $0.15 (Global) |
| Cache storage | $1.00 / 1M tokens / hour |
For comparison, Gemini 3.1 Pro is priced at $2.00/1M input and $12.00/1M output. While Gemini 3.5 Flash is 3x more expensive than its predecessor Gemini 3 Flash ($0.50/$3.00), its performance on agent tasks is significantly improved, and it is faster.
5.2 Antigravity Subscription Plans
Google also adjusted subscription plans at I/O 2026, introducing a new $100 tier:
| Plan | Price | Audience |
|---|---|---|
| Individual | Free | Anyone, with generous weekly rate limits |
| AI Pro | $19.99/month | Developers, more generous rate limits + flexible AI credit pool |
| AI Ultra (New) | $99.99/month | 5x the AI limits of Pro |
| AI Ultra (Previous Top Tier) | $199.99/month (reduced from $249.99) | 20x the AI limits of Pro |
All plans support the following models: Gemini 3.5 Flash, Gemini 3.1 Pro, Gemini 3 Flash, Claude Sonnet & Opus 4.6, gpt-oss-120b.
5.3 Enterprise: Gemini Enterprise Agent Platform
Google Cloud customers can now access Antigravity 2.0 and CLI through the Gemini Enterprise Agent Platform, with:
- Access under Google Cloud Terms of Service
- Google Cloud project integration
- Pay-as-you-go API pricing
- $300 free credits for new customers
5.4 Managed Agents (Gemini API)
Launch an agent with reasoning, tool use, and code execution capabilities through a single API call, running in an isolated Linux environment. During the preview period, environment compute (CPU, memory, sandbox execution) is free, with model inference billed at standard Gemini rates.
6. Impact on Traditional Development
6.1 From "Code Assistance" to "Agent Orchestration"
Traditional AI coding tools (such as GitHub Copilot) are positioned as code completion assistants, with the developer remaining the primary executor. Antigravity 2.0's shift: developers evolve from "people who write code" to "people who manage agents."
Instead of writing code line by line, you now:
- Define goals and constraints
- Launch agents to execute tasks
- Review agent output
- Provide feedback or adjust direction as needed
6.2 Asynchronous Development Becomes the Norm
Scheduled Tasks and asynchronous task management mean developers can assign tasks to agents before leaving work and review results the next morning. Development is no longer a synchronous, continuous-attention process.
6.3 Multi-Model Strategy Becomes Standard
Antigravity supports Gemini, Claude, GPT-OSS, and more, allowing developers to choose the most suitable model for each task's complexity. Simple code completion with Gemini 3.5 Flash, complex reasoning with Claude Opus 4.6 — this flexibility is unavailable with single-model tools.
6.4 Context Expansion: From Repository to Project
Traditional AI coding tools are typically bound to a single repository. Antigravity 2.0's "Project" concept enables agents to work across multiple folders, which is critical for microservices architectures, full-stack projects, and similar scenarios.
6.5 Ecosystem Integration: From Code to Product
Antigravity 2.0's deep integration with Google AI Studio, Android, and Firebase means:
- Prototype in AI Studio, export to Antigravity for local development with one click, deploy to Firebase
- Build Android applications with natural language, publish directly to Google Play testing tracks
- Agents can call Google Workspace APIs and be embedded into applications
7. Ecosystem: SDK, API, and the Broader Vision
7.1 Antigravity SDK
Developers can build custom agents on top of Antigravity, using Markdown files to define custom instructions and skills. Google AI Studio Playground will also provide custom agent templates.
7.2 Google AI Studio Mobile App
The new mobile app allows developers to capture ideas on the go and return to a working prototype by the time they're back at their desk.
7.3 Antigravity in Consumer Products
Google is integrating Antigravity's coding capabilities into consumer products, such as Google Search — where users will see real-time custom UI generated during searches, and can even build mini-applications while exploring topics.
8. Conclusion: The Agent-First Era Has Arrived
The release of Antigravity 2.0 marks Google's definitive judgment on the future of software development: agents are not a feature of the IDE — they are the core of the development process.
From a technical perspective, the shared Agent Harness, co-optimization with Gemini models, multi-agent parallel orchestration, scheduled tasks, and other enterprise-grade capabilities make Antigravity 2.0 far more than an "AI coding assistant" — it is a complete agent development platform.
From a business perspective, the multi-tier pricing from free to $200/month, the introduction of an enterprise plan, and the opening of APIs and SDKs demonstrate Google is building a complete ecosystem around Antigravity.
For developers:
- Individual developers can experience agent-first development with the free plan
- Independent developers and small teams can get higher usage limits through AI Pro ($20/month)
- Heavy users can choose the new $100 tier or the $200 top-tier plan
- Enterprise customers can access pay-as-you-go flexibility and enterprise-grade support through Google Cloud
The retirement of Gemini CLI and the launch of Antigravity CLI further indicate that Google is consolidating all resources into this unified agent platform. After June 18, 2026, Gemini CLI will no longer serve free and Pro/Ultra users — migrating to Antigravity CLI will become the inevitable choice.
The agent-first era of development has arrived.
References:
- Google Antigravity Blog: Introducing Google Antigravity 2.0
- Google Antigravity Blog: Introducing Google Antigravity CLI
- Google I/O 2026 Developer Highlights
- Gemini Developer API Pricing
- Google Cloud Agent Platform Pricing