Input vs Output Cost

Pricing

Part of: Cost & Token Budgeting

Two engineers send the model the same 4,000 tokens. One puts almost all of it in the prompt and asks for a one-word answer. The other sends a tiny prompt and lets the model write a 4,000-token essay. Identical token totals, wildly different bills. The second call can cost five times the first. That gap has a name: asymmetric pricing . What it is LLM pricing is asymmetric : input tokens and output tokens are billed at different rates, and output is almost always the expensive side. A common pattern is something like $3 per million input tokens against $15 per million output tokens, a clean 5x gap. The provider charges more to generate text than to read it. This means the token total alone tells you almost nothing. A call's cost depends on which side the tokens land on. Shift tokens from output to input and the same workload gets cheaper. How it works Cost is two products added together, never one flat per-token rate: Here 3,000 input tokens cost $0.009, but just 800 output tokens cost $0.012, fewer tokens, higher cost, because each output token is priced 5x higher. Why it matters The asymmetry flips your intuition about where to optimize. A bloated prompt feels wasteful, but it's bi

Challenge: The Asymmetry Auditor