How Does OpenRouter Pricing Work? Layers Explained
How does openrouter pricing work? In short, each model has its own price per million input tokens and per million output tokens, set by whichever provider serves it, and you pay from prepaid credit on the router's account. This article lists the layers to check and compares them with a flat rate.
Updated


Key points
- Router bills combine a per-model token price, provider differences and any fees on the router's side; read their current pricing for exact figures.
- Input and output tokens are priced separately everywhere, and output usually costs more.
- A flat model has one input rate and one output rate: $0.15 and $0.50 per 1M tokens here.
- Estimate cost with a short script that reads the usage field of each response.
The pricing layers of a router
A router resells access to models from many providers. That produces a stack of price components. I describe the structure here and leave exact numbers to the router's own pages, because they change and I would be guessing.
- Model token price. Every model lists a rate per 1M input (prompt) tokens and per 1M output (completion) tokens. Larger models cost more than smaller ones, often by a wide margin.
- Provider choice. The same model can be hosted by several providers at different prices, speeds and limits. A router may pick one for you by price, by speed or by availability, so your effective rate can vary.
- Router-side charges. Check whether the router adds a fee when you buy credit, a markup on usage, or minimum purchase amounts.
- Special model variants. Some catalogs list free or discounted variants with tighter limits. Read the limits before building on them.
- Extra features. Anything beyond plain text, such as image input or tool use, may be priced differently by model.
None of this is bad. It is the cost of having a hundred models behind one key. The problem is only that a bill is harder to predict. For a wider comparison see OpenRouter API alternatives compared.
Tokens, input versus output
Whatever the provider, billing counts tokens. A token is a chunk of text, about four characters of English. A prompt of 1,000 words is roughly 1,300 tokens. Input tokens cover everything you send: system prompt, chat history, documents, tool definitions. Output tokens are what the model writes back.
Output usually costs more per token than input, so a chatty answer costs more than a long prompt of the same size. Chat apps also resend history every turn, which means input cost grows through a conversation even when each answer is short.
The exact tokenizer differs by model, so the same text can count as different amounts on different models. When comparing two models by price, compare cost for your own prompt set, not rate cards.
Worked example: the same month, two models
Suppose a workload of 30M input tokens and 6M output tokens a month. Model A on a router has, for this made-up example, a price of $1 and $4 per million input and output tokens. Those figures are made up for the example. The bill is 30 x $1 + 6 x $4 = $30 + $24 = $54, plus whatever fees apply.
On a flat endpoint priced at $0.15 per 1M input and $0.50 per 1M output, the same volume costs 30 x $0.15 + 6 x $0.50 = $4.50 + $3.00 = $7.50. That is not an argument that cheaper always means better. The two models behave differently, and the larger one may be what your task needs. It shows why you should price your actual volume rather than reading headline rates.
Another practical point: many routers show a price for a typical request next to the model. Multiply it by your daily requests and by 30 to sanity-check a plan before committing.
The flat alternative
This site's endpoint, https://api.openrouteralternativeapi.com/v1, has one model, uncensored, and one price list. Input is $0.15 per 1M tokens and output is $0.50 per 1M. You pay from prepaid credit with no subscription, failed or refused requests are not charged, and credit never expires. There is no router layer: the model is run by an independent provider and is not an official product of any big lab.
That makes cost forecasting a one-line calculation. A call with 2,000 input tokens and 500 output tokens costs 2,000 x $0.15 / 1M + 500 x $0.50 / 1M = $0.0003 + $0.00025 = $0.00055. A thousand such calls cost $0.55. If you need many different models, a router is the better fit; if you need one, flat pricing is easier to reason about. See the pricing page for top-up bonuses.
Measure real cost with a tiny script
Do not trust estimates for long. The response includes a usage block with prompt and completion token counts, and this API also returns an X-Balance-USD header with your remaining credit. Use either to log cost per call:
from openai import OpenAI
client = OpenAI(base_url="https://api.openrouteralternativeapi.com/v1", api_key="YOUR_KEY")
PRICE_IN, PRICE_OUT = 0.15, 0.50 # dollars per 1M tokens
def run(prompt):
r = client.chat.completions.create(
model="uncensored",
messages=[{"role": "user", "content": prompt}],
)
u = r.usage
cost = (u.prompt_tokens * PRICE_IN + u.completion_tokens * PRICE_OUT) / 1e6
return r.choices[0].message.content, cost
text, cost = run("List three risks of changing a database schema live.")
print(f"cost this call: ${cost:.6f}")Run it on a hundred real prompts and you will have a cost per task that includes your real prompt lengths. Do the same on a router with a model you are considering, using the numbers on their dashboard, and compare the two totals. The article on the OpenRouter-style OpenAI-compatible endpoint explains how the client side is identical.
Budgeting and top-ups
Set a budget before you start. Prepaid credit helps here, since you cannot spend more than you loaded. On this API, top-ups are any whole amount from $20 to $500, with +5% from $50, +10% from $100 and +15% from $200. A $100 top-up gives $110 of credit. At the $0.00055 per call above, that is 200,000 calls.
You can pay by crypto (USDT on TRC20, USDC on Base). Rate limits protect your budget as well: a full key is capped at 8 concurrent requests and 300 per minute, a trial key at 2 and 60, with HTTP 429 beyond that. Trying it costs nothing: claim the $0.50 trial on the API key page, no card, valid 7 days. Endpoint details are in the docs.
More guides
Questions and answers
Does OpenRouter charge per token?
Yes, each model lists a price per million input tokens and per million output tokens. Check their current pricing page for the exact rates and any fees.
Why can the same model cost different amounts on a router?
Different providers host the same model at different prices, and the router may choose among them. Check how your requests are routed.
How is this API priced?
A flat $0.15 per 1M input tokens and $0.50 per 1M output tokens, from prepaid credit with no subscription.
Am I charged for failed requests here?
No. Failed or refused requests are not charged.