OpenAI Compatible API Pricing Comparison Method

The formula that makes providers comparable

An openai compatible api pricing comparison starts with one fact: almost every provider bills input and output tokens at different rates, quoted per 1M tokens. A single price per million is not enough to rank them. You need your own request shape.

Cost per request = (input tokens ÷ 1,000,000 × input price) + (output tokens ÷ 1,000,000 × output price). Run that for each provider with the same numbers and the comparison is honest. Take the prices from each provider's own pricing page on the day you decide, because they change.

For this service the rates are $0.15 per 1M input tokens and $0.50 per 1M output tokens. Output is about 3.3 times the input rate, so output-heavy workloads cost more per token than prompt-heavy ones.

Three workloads, priced exactly

Here are three common shapes, each computed from the two rates above.

Workload	Input tokens	Output tokens	Input cost	Output cost	Per request	Per 1,000 requests
Short chat turn	2,000	500	$0.00030	$0.00025	$0.00055	$0.55
RAG answer with long context	8,000	300	$0.00120	$0.00015	$0.00135	$1.35
Long story generation	300	2,000	$0.000045	$0.00100	$0.001045	$1.045

Check one row by hand. For the RAG answer: 8,000 ÷ 1,000,000 = 0.008, and 0.008 × $0.15 = $0.0012. Then 300 ÷ 1,000,000 = 0.0003, and 0.0003 × $0.50 = $0.00015. The total is $0.00135.

Notice that the RAG row is the most expensive per request even though it writes the shortest answer. Input volume dominates it. The story row is the opposite: the prompt is tiny and the answer carries nearly all the cost.

Scale it to a month

Per-request numbers are hard to feel. Convert them into a monthly bill using your own traffic estimate.

Say your app sends 50 million input tokens and 10 million output tokens a month. Input: 50 × $0.15 = $7.50. Output: 10 × $0.50 = $5.00. The month costs $12.50.

Now run the same two token counts through any other provider's input and output prices. If their input is cheaper but output is dearer, the answer flips depending on your ratio. This is why a 5:1 input to output ratio and a 1:5 ratio can crown different winners.

A handy shortcut is the blended price: total cost divided by total tokens. Here, $12.50 over 60 million tokens is about $0.21 per 1M tokens. Compare blended prices only when the input to output ratio is the same.

Costs that do not show up in the rate card

• Charges on failures. Here, failed or refused requests are not charged. Ask every provider the same question, because retries on errors add up on a busy app.
• Credit expiry. Credit here never expires, so a slow month costs you nothing extra. Some providers expire prepaid balances; check their terms.
• Minimums and bonuses. Top-ups here are any whole amount from $20 to $500, with +5% from $50, +10% from $100 and +15% from $200.
• Subscriptions. There is no monthly fee here. A flat fee changes the math for low-volume users.
• Payment method. Payment here is crypto only, in USDT on TRC20 or USDC on Base. If you cannot pay that way, the cheapest rate does not help you.
• Context limits. The window is 100,000 tokens. A provider with a smaller window may force you to chunk long inputs, which means more calls.

A worked comparison sheet you can copy

Put the numbers in one small table and fill it in once per provider. Keep the workload constant and change only the prices. The columns that matter are: input price per 1M, output price per 1M, cost of your short turn, cost of your long-context call, charges on failed requests, credit expiry and minimum deposit.

For this service the row reads: $0.15, $0.50, $0.00055, $0.00135, none charged, never expires, $20 minimum top-up. Fill the same row from each competitor's own pages, and do it on the day you decide.

Add one more calculation that people skip: growth over a chat. Because every turn resends the history, a conversation of 10 turns, where the history grows by 400 tokens per turn from a 300-token first prompt, sends 21,000 input tokens in total (300 + 700 + 1,100 and so on). At $0.15 per 1M that input is $0.00315. Two providers with the same input price will produce the same input cost here, so the difference then comes only from output price and from how many tokens the model chooses to write.

That last point is easy to overlook. A model that writes shorter answers for the same task can cost less per task even at a higher per-token price. Measure output length on your own prompts, not only the price list.

• Run 50 identical prompts through each candidate.
• Record average input and output tokens from the usage field.
• Multiply by each rate card.
• Compare quality by reading a sample, not by counting tokens.

What the bonuses do to the effective price

The top-up bonus is extra credit, so it lowers your effective rate. A $200 top-up gives you $230 of credit, because 15% of $200 is $30. Your effective prices are then $0.15 ÷ 1.15, about $0.130 per 1M input tokens, and $0.50 ÷ 1.15, about $0.435 per 1M output tokens.

Top-up	Bonus	Credit received
$20	none	$20.00
$50	+5%	$52.50
$100	+10%	$110.00
$200	+15%	$230.00
$500	+15%	$575.00

The same short chat turn that costs $0.00055 at list price would cost about $0.00048 with the 15% bonus. Whether a bigger top-up is worth it depends on how fast you will spend it. Since credit never expires, the main risk is only tying up cash.

A short checklist for your own comparison

1. Measure real token counts from a sample of 100 to 200 of your requests, split into input and output.
2. Compute the three numbers above for each provider: per request, per 1,000 requests and per month.
3. Add the non-rate costs: failed-call charges, expiry, minimums and fees.
4. Test quality on your own prompts. A model that costs half as much but needs two attempts is not cheaper.
5. Check that the API really is compatible with your client by running the steps in the docs.

To try this service at no cost, claim the $0.50 trial credit on the key page. At the short chat rate of $0.00055 per request, that is about 900 calls. Current rates and bonuses are always on the pricing page.

More guides

• Open AI Compatible Uncensored API: What Works
• How to integrate an open router alternative