Top AI Product

systima.ai wired a logging proxy between two AI coding agents and the API, capturing every request byte-for-byte. Claude Code ships ~33,000 tokens before your first word arrives: a 6.5k system prompt, 24k of schemas for 27 tools, 2k of message scaffolding. OpenCode, with 10 tools, ships ~7k. That’s 4.7x on Sonnet 4.5, narrowing to 3.3x on Fable 5.

Real configs are worse

Add a CLAUDE.md and a few MCP servers and the baseline hits 75,000–90,000 tokens — 40% of the 200k context window gone before you type anything. Fan a small task out to two subagents and 121k tokens becomes 513k, a 4.2x multiplier. Claude Code also rewrites its prompt prefix mid-session, writing up to 54x more cache tokens than OpenCode, and cache writes bill at a premium.

Same output, different bill

The damning part: on every test task, both harnesses produced identical correct results. The extra tokens bought nothing. That’s why HN gave it 329 points and 185 comments — every Claude Code user just saw their invisible daily surcharge quantified, with a receipt.

You Might Also Like

• Pi Mono 29k Stars and a 200 Token System Prompt That Rivals Claude Code
• Claude Code Security Just Dropped and it Already Found 500 Zero Days Nobody Knew About
• Claude Code Remote Control Just Turned my Phone Into a Coding Terminal and im Weirdly Into it
• Obra Superpowers Turned my Claude Code Into a Proper Engineer and im not Going Back
• Claude Context Mode Might be the Best Thing Thats Happened to my Claude Code Sessions