Antigravity 2.0 Looks Great On Paper, But Claude Code & Codex Are Not Going Anywhere

Google's Antigravity 2.0 hits the market faster than Claude Code, but speed alone has never been what keeps developers loyal. Here is what the new tool gets right and what it still gets wrong.

Announced at Google I/O 2026, Antigravity 2.0 runs on Gemini 3.5 Flash and claims record coding speeds.

Source : Google

Show Quick Read

Key points generated by AI, verified by newsroom

• Google launched Antigravity 2.0 as a standalone desktop coding tool.
• It uses the new Gemini 3.5 Flash model for enhanced speed.
• Key features include parallel subagent execution and asynchronous task handling.

Google announced Antigravity 2.0 at I/O 2026, rebuilding its coding tool from the ground up as a standalone desktop app rather than an IDE with AI features added on. Available now on macOS, Linux, and Windows, it runs on a new model called Gemini 3.5 Flash. 

The launch is Google's most direct challenge yet to Claude Code and OpenAI Codex, two tools that have built real, lasting followings among developers over the past year. Speed is the headline, but speed alone has never been what keeps developers loyal to a tool.

Why Raw Speed Will Not Be Enough To Pull Developers Away

Google's big number is 289 output tokens per second on Artificial Analysis, against 67 for Claude Opus 4.7 and 71 for GPT-5.5. Google also claims Gemini 3.5 Flash beats its own Gemini 3.1 Pro on coding and agentic benchmarks. 

Those figures are notable, but benchmark performance rarely reflects what daily use actually feels like. Claude Code and OpenAI Codex have months of developer trust built around them, and that is not something a faster model overrides overnight.

Antigravity head Varun Mohan made the case on stage, saying the agents built a full operating system from scratch for under $1,000 in tokens, then ran a Doom clone on top of it.

What Is Actually New Inside Antigravity 2.0

The core change is structural. The main agent can now spin off subagents on the fly, run them in parallel, and keep its own context window clean. Tasks run asynchronously, so nothing stalls while something compiles. JSON hooks let developers shape agent behaviour without writing wrappers. A Scheduled Tasks layer adds cron support, so agents can fire on a timer.

New slash commands shape how the tool is used: /goal runs to completion without checking in, /grill-me makes the agent ask questions before touching a file, and /browser is a manual on-switch because, as Google concedes, its agents still cannot reliably decide when to open a webpage on their own. Voice input now transcribes live, and conversations sit inside "projects" that span folders with their own permissions.

On pricing, a new $100 AI Ultra tier sits between the $20 Pro plan and the top option, offering five times the Antigravity limits of Pro. The flagship Ultra drops from $250 to $200, with 20 times the Pro limits.

New and existing Ultra subscribers get $100 in bonus credits through May 25. A new Antigravity CLI replaces the Gemini CLI entirely, meaning existing terminal workflows will need to be migrated. Existing Antigravity IDE users roll over to 2.0 on their next update.

Google Releases Android 17 QPR1 Beta 3 With Seven Bug Fixes For Pixel