The 'Dynamic Languages Are More Token-Efficient' Claim Falls Apart Under Real-World Tests
What's the best programming language for coding agents?
A widely cited eval claims dynamic languages like Clojure are far more token-efficient for LLM coding than static ones like Rust. But Dan Luu's own tests on zstd and Pandoc show the advantage vanishes at higher effort levels, and obscure dense languages like J perform poorly. Popularity correlates with better results, suggesting mainstream languages are the safer bet for LLM users.
Perhaps using an obscure (and "weird") language can make sense if you have a very large budget and you can train or fine-tune a model to be effective for your pet language, but if you're a normal user of LLMs, it seems like sticking with a mainstream language is likely a better bet than using an obscure dense language.