Yesterday I wrote 2 more specs for my program, and needed a LLM to generate the code. I used OpenRouter and decided to try out GLM 5.3 Flash. My total bill for the endeavor came out to about 3 cents ($0.03).
After using GLM 5.3 Flash I can see why some people want to buy expensive computers and run it locally all day—it is a nice model. It coded the functions well, and worked quickly. Both of the Rust functions were optimized in ways better than the original versions.
Some notes on spec-driven development: • There is a "complexity level" that is ideal.
• Writing a complete spec for a simple function is kind of a waste of time; and code that is too complex also needs more than a single specification.
• Using a handful of specs and generating code for them via chat on OpenRouter is surprisingly cost-efficient, even with models like GPT-6-Sol.