3 min read
Developer traces 3,177 API calls to compare 4 AI coding tools' context usage
A developer built Context Lens to trace LLM context usage. Testing four AI coding tools on the same bug fix revealed vastly different token consumption and strategies: Claude Opus was surgical (23K tokens), Codex efficient (29-47K), Claude Sonnet balanced (42-44K), and Gemini profligate (up to 350K tokens). The tools show no deliberate context management; efficiency stems from investigation strategy, not optimization.