Test: Antigravity CLI uses 68–88% fewer tokens than Claude Code

A developer benchmarked Antigravity CLI, Claude Code and Codex on identical prompts across 44 runs. With the same Claude Opus 5.5 model, Antigravity read 68–88% fewer input tokens, wrote 40% fewer output tokens and finished 27% faster. All 12 generated games worked in every tool.

• Antigravity + Opus 5.5: 9,822 input tokens vs 78,671 for Claude Code
• On the "OK" prompt Claude Code used 35,781 tokens, Antigravity 5,326
• Claude Code produced longer code: 451 lines vs 236 for Antigravity
• 30 MCP tools added 388 tokens to Claude Code and 0 to Codex