Claude Code sends 33k tokens before reading the prompt; OpenCode sends 7k

A comparative analysis of Claude Code and OpenCode found that Claude Code consumes significantly more tokens before processing user requests—33,000 tokens versus OpenCode's 7,000—due to larger system prompts and tool schemas, and is also far less efficient with prompt caching, writing cache tokens at premium rates up to 54x more than OpenCode. Production setups further bloat costs with configuration files and MCP servers adding 25,000-92,000 tokens before any user input, while subagent routing can multiply token consumption by 4x due to bootstrap costs and transcript overhead. The sole advantage for Claude Code is batch processing of tool calls on multi-step tasks, which can reduce total session costs compared to OpenCode's higher per-turn baseline.