Question 1

How is the monthly cost for each provider calculated?

Accepted Answer

Each provider's monthly cost is (input tokens per request ÷ 1,000,000 × input price per 1M) + (output tokens per request ÷ 1,000,000 × output price per 1M), all multiplied by your monthly request volume. LLM APIs price input and output tokens separately, so the formula keeps them separate. Example: 1,000 input tokens and 500 output tokens at $3/M input and $15/M output is (1,000/1,000,000 × $3) + (500/1,000,000 × $15) = $0.003 + $0.0075 = $0.0105 per request. At 10,000 requests/month that is $105/month. The comparator runs this for every provider you enter and ranks them cheapest first.

Question 2

Where do I find the current price for each provider?

Accepted Answer

Each major provider publishes a pricing page in dollars per 1,000,000 tokens, split by model and by input vs output: Anthropic at anthropic.com/pricing, OpenAI at openai.com/api/pricing, and Google Gemini at ai.google.dev/pricing. Open the page for the specific model you plan to use, copy the input and output per-1M prices, and enter them here. The comparator never hardcodes a current price — the example providers are seeded with dated illustrative numbers and a 'verify current pricing' note precisely because provider prices change often. Always confirm the live number before you rely on the result.

Question 3

Why does output length change which provider wins?

Accepted Answer

Output tokens are consistently 3-5x more expensive than input tokens across every major provider, because generating each output token requires a full forward pass through the model while input tokens are processed in parallel. That means the ranking is driven by your output:input ratio, not just total volume. A provider with cheap input but expensive output can win a classification workload (short outputs) and lose a long-form generation workload (long outputs). When you change the output-tokens-per-request input above, watch the ranking reshuffle — that is the whole point of comparing on your actual profile rather than a headline per-token price.

Question 4

Does this include infrastructure, self-hosting, or GPU costs?

Accepted Answer

No. This compares token-metered API pricing only. Self-hosting an open-source model (Llama, Mistral, Mixtral) on dedicated GPU instances has a completely different cost structure — you pay for GPU-hours whether or not you send requests, plus engineering overhead for model serving, scaling, updates, and reliability. Self-hosting can beat API pricing at very high, steady volume and lose badly at low or spiky volume. To compare self-hosting, model the fully-loaded GPU cost separately and enter it as an effective per-1M-token price, but treat that number as an estimate, not a published rate.

Question 5

How does prompt caching affect the comparison?

Accepted Answer

Prompt caching can change the ranking entirely, and this tool does not model it for you. Caching stores the processed prefix of a repeated prompt so subsequent calls pay a steep discount on the cached portion — Anthropic's cache reads are roughly 10% of the standard input rate, for example. If your workload has a large, stable system prompt reused across many calls, your effective input price is far below the headline rate, and a provider that offers good caching can leapfrog a nominally cheaper one. To approximate it here, compute your blended effective input price (cache-write on the first call, cache-read on the rest) and enter that as the input price for the caching provider. See the Prompt Caching Savings Calculator to work out that blended rate first.

Question 6

What about batch discounts and committed-use pricing?

Accepted Answer

Several providers offer a discounted batch tier (often around 50% off) for non-real-time work, plus committed-use or enterprise pricing negotiated with their sales team. This comparator uses whatever per-1M prices you enter, so to model a batch or committed rate, simply enter the discounted price for that provider's row. Just be consistent: if you compare one provider's batch price against another's real-time price, the comparison is no longer apples-to-apples. Decide whether your workload is batchable first, then compare like tiers against like tiers.

Question 7

Can I compare more than three providers?

Accepted Answer

Yes. The provider list is fully editable — add a row for every model and provider you are weighing, give each a name and its input and output per-1M prices, and the table and chart update to rank all of them. A common pattern is to enter the same provider twice at two different models (a frontier model and a mid-tier model) to see exactly how much you would save by dropping to the cheaper model for part of your workload. There is no provider built in; you decide which ones appear.

Question 8

Does a cheaper provider mean I should switch?

Accepted Answer

Not on its own. This tool reports the arithmetic: at the inputs you entered, it tells you which provider is cheapest and by how much. It does not weigh model quality, latency, rate limits, context-window size, multi-modal support, data-residency, or reliability — all of which can matter more than price for a given product. The honest decision is: compute the cost spread here, then judge whether the cheaper option clears your quality and operational bar on your own evaluation set. The number is one input to the decision, not the decision.

Provider	Per request	Monthly	vs cheapest	Saves vs max
Gemini 1.5 Flashcheapest	$0.000225	$22.50	1.00×	$1,027.50
GPT-4o mini	$0.000450	$45.00	2.00×	$1,005.00
Claude 3.5 Sonnet	$0.010500	$1,050.00	46.67×	—

LLM Provider Cost Comparator

What this means

Worked example

Frequently asked questions

Show the math