LLM APIs vary 250x, agents balloon token use

- Analysts flagged a 250x spread in LLM API pricing across vendors, citing examples from low‑cost lfm2‑8b to expensive Claude Opus lines while agentic flows multiply token consumption. (x.com) - Example pricing given: liquid’s lfm2‑8b around $0.01 per thousand input tokens versus Claude Opus near $5 per thousand, with agentic workflows consuming 10–20× more tokens. (x.com) - That cost divergence makes new decoding methods like wavefront sampling (a staggered‑token idea claiming ~5× cheaper decoding) strategically important for expensive agent deployments. (x.com) (x.com)