🤖RTK Filters Terminal Output, Saves 89% Tokens But Costs More
RTK cuts tokens but costs more for some tasks
TL;DR
RTK filters and compresses terminal output, saving 89% tokens but increasing costs for some tasks. Fable saw 5% cost savings, while DeepSeek costs rose by 5%. Pass rates dropped slightly.
RTK filters and compresses terminal output before the AI agent reads it, saving 89% of tokens. However, this doesn't always translate to cost savings. In a benchmark test, Fable tasks saw a 5% cost reduction, while DeepSeek tasks were 5% more expensive. Pass rates also dropped by 1% for Fable and 2% for DeepSeek. RTK's impact varies widely: one Fable task saw 95% cost savings, while DeepSeek tasks often took more turns, increasing costs. The real story is that RTK doesn't guarantee cheaper AI coding, despite significant token savings.
Key Points
RTK filters and compresses terminal output, saving 349.2 million tokens, a 89% reduction.
Fable tasks saw a 5% cost reduction, while DeepSeek tasks were 5% more expensive.
Pass rates dropped by 1% for Fable and 2% for DeepSeek.
One Fable task saw 95% cost savings, while DeepSeek tasks often took more turns, increasing costs.
RTK filters only shell commands, reducing terminal-output characters by 9%, yet prompt tokens rose 9%.
Why It Matters
If you're using RTK to filter terminal output, be aware that while it can save 89% of tokens, it doesn't necessarily reduce costs. Fable tasks saw a 5% cost reduction, but DeepSeek tasks were 5% more expensive. Pass rates also dropped slightly. RTK's impact varies widely, so monitor costs closely. The real story is that RTK doesn't guarantee cheaper AI coding, despite significant token savings.
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 3,476 builders reading daily.