Systima Technologies conducted an analysis revealing significant differences in the token overhead of AI coding agents. Claude Code, using Claude Sonnet 4.5, incurred approximately 33,000 tokens of fixed overhead, including system prompts and tool schemas, before processing a user's prompt. In contrast, OpenCode, using OpenCode 1.17.18, required only about 7,000 tokens for the same task. This substantial difference in token usage, which directly impacts cost, was attributed to Claude Code's tendency to re-write identical cache prefixes multiple times, unlike OpenCode which maintained a stable cache. The study also noted that Claude Code silently ignored an instruction file unless renamed, and that sub-agent usage dramatically increases token consumption. AI
IMPACT Highlights efficiency as a key competitive factor for AI coding agents, potentially influencing future development and cost structures.
RANK_REASON Analysis of AI agent efficiency and cost implications.
Read on dev.to — Claude Code tag →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
