How to cut your AI API bill
Five practical ways to spend less on model APIs without guessing. New to the numbers? Start with how AI pricing works or the pricing glossary.
Match the model to the job
Do not pay frontier prices for simple chat. Our cheapest-for-each-job pages rank models that can handle chat, coding, long documents or hard reasoning. Today's lowest list input price is $0.075/1M for GPT-OSS 20B (Groq) — but cheap is only useful if it still does your work.
Watch output tokens — they cost more
Most providers charge several times more for the text the model writes than for the text you send. Ask for shorter answers, use bullet lists, and avoid “write a long essay” when a paragraph will do. See how input vs output pricing works.
Shorten what you send in
Repeated instructions, long pasted docs and chat history all add input tokens. Keep system prompts tight, drop unused context, and prefer provider “cached input” or batch discounts when they fit (our table shows the standard list rate only).
Compare before you switch
A lower sticker price can still lose if your mix of input and output is unusual. Use a head-to-head compare for the two models you are considering, or plug your own volumes into the savings calculator.
Re-check prices — they move
Providers cut and raise rates. Follow the price-change log or the RSS feed, watch for sale prices that are about to end on our deals page, and always open the official pricing page linked from each model before you buy.
Browse the full price table, see example monthly workloads, pick by provider or model, or estimate a switch in the savings calculator.
Tips use standard pay-as-you-go text prices. Batch, cached-input and long-context rates differ — check the provider’s own page.