RTK reports token savings, but our cost benchmarks disagree

RTK, a popular tool claiming to reduce AI coding costs by filtering terminal output, showed minimal or negative cost savings in benchmark testing. When researchers tested RTK on Terminal-Bench 2.1 using Claude Code and OpenCode across 1,740 attempts, they found costs decreased by only 3% for Fable while increasing 7% for DeepSeek, with slight reductions in pass rates for both models. The findings suggest that while RTK compresses terminal output effectively, this doesn't reliably translate to cheaper AI coding, contradicting marketing claims of up to 60% token savings.