RTK

Reading less output does not equal spending less money; the true cost of AI agents is hidden in context caching, extra rounds, and the blowback caused by task quality fluctuations—none of which can be compressed by a compression tool.

The RTK tool summarizes terminal output to reduce what Claude reads, but JetBrains measured across hundreds of tasks and found the bill was essentially the same or even slightly higher

After installing RTK, Claude's task pass rate dropped by 1 percentage point and DeepSeek's dropped by 2%.

In task-level cost accounting, DeepSeek's per-task cost surged by 17%, and looking only at the 36 fully passed tasks, the cost still increased by 18%.

RTK's core metric, RTK gain, is calculated as the raw output minus the filtered output divided by 4, counting only the deleted tokens that are not billed at all, without accounting for subsequent round costs.

The disclaimer in RTK's Readme states that it reduces tool output by up to 90%, but this does not equate to a 90% reduction in the bill.

Without RTK, tool output accounts for only 7% of non-model output, while cache reads account for 94% to 98% of total token usage.

RTK compressed tool characters by 9%, but prompt tokens actually increased by 9%, with average per-round input down 7% but total rounds up 18%.

The test ultimately produced a sample of 174 tasks and 1,744 attempts, using an A/B comparison method with the same routing, same platform, and same time period.

On the Claude line, the total bill without RTK was $731 and with RTK was $698, a reduction of approximately 5%; on the DeepSeek line, without RTK the cost was $51, but with RTK installed it actually rose to $54, a cost increase of 5%.

RTK publicly claims to significantly reduce token consumption in large language model calls, thereby helping users save on API costs.

The token savings counted internally by the tool may be based on a theoretical baseline rather than the user's actual call patterns, and the higher the baseline is set, the better the savings figure appears.

RTK (Rust Token Killer) is an experimental output compression tool written in Rust; Rust was chosen for its extremely high execution performance and memory safety guarantees.

A hands-on blog post tested RTK by spending over $1,500 and running more than 1,700 tasks.

During testing, a bug triggered by RTK compression caused an infinite loop, with 339 consecutive errors stalling for 12 minutes before timing out, and the single-task cost spiked to 9 times the baseline.