AI API Cost Calculator

Almost every language model API bills by the token, not by the request. A token is a chunk of text — roughly four characters, or about three-quarters of an English word — and both the prompt you send and the completion the model returns are counted. Prices are quoted per million tokens, which makes the numbers look trivially small: $2.50 per million sounds like nothing until you notice a single retrieval-augmented chat turn can carry 20,000 tokens of context, and that you are serving it tens of thousands of times a day. This calculator exists to turn that per-million rate into the two figures that actually drive a budget — cost per request, and cost across your real monthly volume — and to line every model up in one sorted table so the cheap option for your specific token mix is obvious rather than buried in seven different pricing pages.

Not every modality is a token. Image APIs mostly charge a flat rate per generated image — FLUX.1 Schnell at about $0.003, Gemini 2.5 Flash Image at $0.039, Ideogram v3 at $0.09 — a 30× spread that matters enormously at volume. Video is billed per second of output, from Veo 3.1 Lite at $0.05 to Veo 3.1 Standard at $0.40, so a single 60-second clip ranges from $3 to $24 depending only on the model. A few image models, such as OpenAI's GPT Image 2, are priced by image tokens rather than per image; because this tool has no image-token input to price them honestly, they are deliberately left out of the per-image comparison rather than shown as a misleading zero. Embeddings are the cheapest line on any AI bill — OpenAI's text-embedding-3-small is $0.02 per million tokens — which is why embedding a whole corpus for search usually costs less than a single day of chat traffic.

The cheapest model is rarely the right default and the flagship is rarely necessary. The highest-leverage move is to route by difficulty: send classification, extraction, and simple summarisation to a mini or nano tier, Gemini Flash-Lite, or DeepSeek, and reserve GPT-5.5, Claude Opus, or Gemini Pro for genuine reasoning and high-stakes output. Because the gap between tiers is often 10× to 50×, shifting routine traffic downward usually saves more than any other single change. Estimate before you build: put your realistic average token counts and monthly volume into the calculator, compare the top few candidates, and you will frequently find two models of similar quality that differ several-fold in cost. What this tool deliberately does not model is everything beyond the base per-token rate — batch tiers (commonly around 50% off), regional and data-residency endpoints, fine-tuning surcharges, and server-side tools like web search all change the final invoice, and every provider's live pricing page is linked in the tool so you can confirm the current figure before you commit budget. The prices here were gathered in July 2026 and are a planning aid, not a contract.