Cutting LLM Token Costs with rtk, headroom, and caveman - savings measured on real workloads
A Reddit user analyzed token cost savings from rtk, headroom, and caveman on $926 worth of real Claude Code workloads, totaling 614M tokens. The user replayed each optimization method over 500 sessions, measuring actual bill reductions. Savings ranged from 60-90% as reported, but varied by method: headroom was directly tested with a pure function compressor, while rtk and caveman were estimated from published ratios. You can apply similar analysis to your own usage patterns.
Key takeaways
- 614M tokens, $926 baseline spend, 60-90% reported savings.
- headroom directly tested; rtk, caveman estimated from published data.
- Actual savings vary by optimization method and usage patterns.
Cutting LLM Token Costs with rtk, headroom, and caveman - savings measured on real workloads
A Reddit user analyzed token cost savings from rtk, headroom, and caveman on $926 worth of real Claude Code workloads, totaling 614M tokens. The user replayed each optimization method over 500 sessions, measuring actual bill reductions. Savings ranged from 60-90% as reported, but varied by method: headroom was directly tested with a pure function compressor, while rtk and caveman were estimated from published ratios. You can apply similar analysis to your own usage patterns.
Key takeaways
- 614M tokens, $926 baseline spend, 60-90% reported savings.
- headroom directly tested; rtk, caveman estimated from published data.
- Actual savings vary by optimization method and usage patterns.