What I Learned From My Claude Code Dashboard

A developer shares insights from setting up a Claude Code telemetry dashboard using BetterStack and OpenTelemetry. Key findings include: Anthropic is heavily subsidizing Claude Code (API costs are roughly 9x the subscription price), subagents silently default to cheaper models like Haiku and Sonnet, token volumes are dominated by prompt cache reads (105M reads vs 3M writes on active days), and the Claude Code team shipped 23 releases in the tracking period. The dashboard was built by connecting the BetterStack MCP server to Claude Code itself, which queried schemas and wrote SQL for each panel.