Show HN: Does per-step reasoning effort save money in Claude Code?

Category: ai-ml

Tags: llm, cost-optimization, claude-code, proxy, reasoning-effort

Score: 6.3/10 (Innovation: 6, Technical: 7, Documentation: 8, Utility: 4)

jev-effort is a research proxy for Claude Code that dynamically adjusts reasoning effort per step using a small decision model, while preserving the prompt cache. It measures cost savings from this approach, finding negligible savings at 'high' effort but up to 55% at 'max' effort in a benchmark, though savings in long real sessions are only 4-9% due to context re-reading dominating costs.

Target audience: AI developers and engineers using Claude Code, particularly those concerned with cost optimization

Repository: https://github.com/ifoster01/jev-effort · JavaScript · MIT · 4 stars

View on Hacker News