A dollar goes a lot further on cheap models
Thinking in tokens-per-dollar makes model choice obvious: for bulk, non-critical work a small model buys 10–20× the tokens. Compare per-call on the LLM token cost calculator.
Tokens per dollar: the cross-model shopping unit
Inverting price-per-million-tokens into tokens-per-dollar makes model shopping concrete: at $0.15 per million input tokens, a dollar buys 6.7M tokens — roughly 5M words, or fifty novels of context. At $15 per million, the same dollar buys 67k tokens. The 100× spread between frontier and budget models is the entire cost-optimization opportunity in most LLM apps: routing the 80% of easy queries to a cheap model and reserving the expensive one for hard cases cuts blended cost 5–10× with minimal quality loss. Output tokens price 2–5× input on most providers — the ratio matters more than either number for chat-heavy workloads.