TL;DR: A head-to-head measurement shows Claude Code uses significantly more tokens (33k) than OpenCode (7k) before reading the prompt, highlighting a major cache inefficiency for AI builders.
Summary: Developers compared token usage between agentic coding tools Claude Code and OpenCode by adding logging to Anthropic's endpoint. Claude Code consistently sent ~33k tokens per task before processing the prompt, while OpenCode sent only ~7k tokens, indicating a suboptimal cache strategy in Claude Code. The finding has sparked discussion about cost and efficiency in agentic coding workflows.
Why it matters: This inefficiency directly increases API costs for teams using Claude Code compared to OpenCode. Builders should audit their agentic tools' token consumption and consider switching or optimizing prompt caching strategies.
Source: rss