OpenRouter pricing: the 5.5% fee, free models, and what routing costs

OpenRouter charges nothing on inference and 5.5% when you buy credits, with a $0.80 minimum. That is the whole pricing model. The arithmetic that matters is what the minimum does to small top-ups and what the endpoint spread does to the price on the model card.

By the Continuum team. We build a workbench that runs Claude Code, Codex, and their peers, so the model rates quoted here are the ones our own cost analytics ship with.

The short version

OpenRouter adds no markup to model tokens. It charges 5.5% of the order with a $0.80 minimum on card top-ups, or 5.0% flat with no minimum on crypto, added on top of the purchase. A $100 credit purchase therefore costs $105.50 and buys $100 of inference at provider list price. Free models are capped at 20 requests per minute and 50 per day, rising to 1,000 per day after a lifetime $10 credit purchase. BYOK is free up to $25,000 per month of list-price inference, then 5%.

What you need to know
  • Card top-up: 5.5% of the order, $0.80 minimum. Crypto: 5.0% flat, no minimum.
  • The fee is added on top: $100 of credits costs $105.50 and buys $100 of tokens.
  • At a $5 top-up the minimum makes the effective fee 16%. Top up in bigger blocks.
  • Zero markup on inference. Token rates match provider list prices.
  • BYOK: free below $25,000/month of list-price inference, then 5% charged to credits.
  • The model card price is the cheapest endpoint. Pin providers if you need that number to hold.

The fee, exactly

OpenRouter simplified its fee on 9 June 2025, removing a fixed $0.35 Stripe charge in favour of a single percentage. The structure has been stable since and is what the pricing page shows in August 2026.

Every charge OpenRouter levies.checked aug 2026
ChargeRateNotes
Credit purchase, card5.5% of order$0.80 minimum. Added on top.
Credit purchase, crypto5.0% of orderNo minimum. Added on top.
Inference$0 markupProvider list price, passed through
BYOK5% of equivalent list priceOnly above the monthly allowance
SubscriptionNoneNo seats, no platform monthly
Logging opt-inminus 1%A discount, not a charge
What a top-up actually costs. OpenRouter's own published comparison for the June 2025 change is in the last column.
Credits boughtFeeYou payEffective overhead
$5$0.80 (minimum)$5.8016.0%
$10$0.80 (minimum)$10.808.0%
$25$1.38$26.385.5%
$100$5.50$105.505.5%
$1,000$55.00$1,055.005.5%
$100 via crypto$5.00$105.005.0%

Direct versus OpenRouter, five models

Because there is no inference markup, the per-token rates are identical. The difference is entirely the 5.5% on the way in, which makes the interesting question which model to spend it on rather than where to buy the tokens; our model directory keeps a card per model with the independently published coding scores next to the list price. Prices below are per million tokens, read from OpenRouter's models API on 19 August 2026.

Cost to run 10M input and 2M output tokens. Direct column is the provider's published rate; OpenRouter column is the same spend grossed up by the 5.5% card fee.
ModelIn / Out per MtokDirectVia OpenRouter
anthropic/claude-sonnet-5$2.00 / $10.00$40.00$42.20
anthropic/claude-opus-5$5.00 / $25.00$100.00$105.50
openai/gpt-5.6-luna$0.20 / $1.20$4.40$4.64
x-ai/grok-4.6$2.00 / $6.00$32.00$33.76
deepseek/deepseek-v4-pro$1.44 / $2.88$20.16$21.27

That is the honest answer to "how much does OpenRouter cost": 5.5% of whatever you were going to spend anyway. On $40 of Sonnet it is $2.20. On a $10,000 monthly bill it is $550, which is the point where a team should be asking whether a zero-fee gateway is worth the integration work.

The one place the model card lies to you

openai/gpt-5.6-sol lists at $2.50 input and $15.00 output. OpenAI's published rate for the same model is $5.00 and $30.00 at short context. Both are true: OpenRouter lists the cheapest of seven endpoints, and it passes each through without markup.

Endpoints serving openai/gpt-5.6-sol, from the endpoints API, 19 August 2026.
ProviderInputOutputvs listed price
OpenAI$2.50$15.00the listed price
OpenAI (batch)$1.25$7.50half
OpenAI (standard)$5.00$30.002x
Azure$5.00$30.002x
Azure$5.50$33.002.2x
Amazon Bedrock$5.50$33.002.2x

Free models: quotas, not pricing

Seventeen model ids ended in :free on 19 August 2026. They cost nothing per token. The constraint is throughput.

Free-model rate limits, from the API reference.checked aug 2026
Lifetime credits purchasedRequests/minRequests/day
$02050
$10 or more201,000

The $10 is a lifetime purchase, not a balance. Buy $10 of credits once, spend them, and the 1,000/day cap stays. The 20 RPM ceiling does not move at any tier.

The free catalogue skews to open-weight models: NVIDIA Nemotron 3 variants, Google Gemma 4, Z.ai GLM 5.2, OpenAI gpt-oss-20b, Cohere North Mini Code, and a rotating set of previews. Frontier models are never free. OpenRouter's docs say free variants are "usually not suitable for production use", and 50 requests a day on a fresh account makes that plain.

BYOK pricing

Attaching your own provider credentials keeps your negotiated rates and provider-side quotas while still routing through OpenRouter.

BYOK fee structure, from the BYOK docs.
TierFree allowance per monthFee above it
Pay-as-you-go$25,000 of list-price inference5%
Enterprise$200,000 of list-price inference5%

The allowance is measured in list-price inference value, not request count, and the 5% is charged against your OpenRouter credit balance rather than your provider bill. Two behaviours to know: BYOK endpoints are always attempted first and bypass your provider ordering, and BYOK spend does not count against guardrail or workspace budgets unless you turn that on.

Getting your bill down

  1. Stop making small top-ups. The $0.80 minimum is the single biggest avoidable cost on a hobby account. $14.55 is break-even.
  2. Pay in crypto if you can. 5.0% with no minimum against 5.5% with one.
  3. Pin the cheap endpoint. :floor or sort: "price". Check the endpoints API before you budget.
  4. Use :batch variants for anything asynchronous. Half price on every model that offers one.
  5. Take the 1% logging discount only if your prompts genuinely carry nothing sensitive. It is 1%, which is less than the top-up fee.
  6. Check whether you need the aggregator at all. If you call one model from one vendor, 5.5% buys you nothing.

For coding agents specifically, the biggest lever is usually not the gateway fee at all: it is context discipline and model choice per task. Reducing Claude Code costs has the arithmetic.

Questions people ask

How much does OpenRouter cost?

Nothing per month and nothing extra per token. You pay a 5.5% fee with a $0.80 minimum when you buy credits with a card, or 5.0% flat with crypto, added on top of the purchase. A $100 top-up costs $105.50 and buys $100 of inference at provider list prices.

Does OpenRouter mark up model prices?

No. Its FAQ states that it passes through provider pricing without any markup. The only charge is on credit purchases. Note that the price shown on a model card is the cheapest of several endpoints, so the rate you actually pay depends on where your request routes.

What is the OpenRouter minimum fee?

$0.80 per credit purchase on cards. That makes small top-ups expensive: a $5 purchase carries a 16% effective fee, and break-even against the flat 5.5% is $14.55. Crypto purchases are 5.0% with no minimum.

How many free requests does OpenRouter give you per day?

50 requests per day on models whose id ends in :free, if you have never purchased credits. Once you have bought at least $10 of credits at any point, the daily cap rises to 1,000. Both tiers are capped at 20 requests per minute.

How much does OpenRouter BYOK cost?

Nothing up to $25,000 of list-price inference per month on pay-as-you-go, or $200,000 on enterprise. Above the allowance it is 5% of what the same model and provider would have cost at list price, deducted from your OpenRouter credits.

Is OpenRouter cheaper than using the provider directly?

No, it is about 5.5% more expensive on the same tokens, because the token rates are identical and the credit fee is additive. It can be cheaper in practice if it routes you to a discounted endpoint you would not otherwise have access to, or if it saves you from paying for capacity at multiple vendors.

Sources

Every figure above was read from these pages on August 2026. Vendors reprice without notice; if you find a stale number, tell us.

  1. OpenRouter pricing 5.5% platform fee, BYOK allowances
  2. OpenRouter FAQ no-markup statement, crypto fee, logging discount
  3. Simplifying our platform fee 9 June 2025 fee change and worked examples
  4. OpenRouter API rate limits free-model RPM and RPD thresholds
  5. OpenRouter BYOK $25,000 allowance, 5% fee mechanics
  6. OpenRouter models API per-model rates read 19 Aug 2026
  7. OpenAI API pricing GPT-5.6 Sol/Terra/Luna published rates
Try it

Metered credits,
or a flat invoice.

Continuum hosted inference is a monthly plan with a weekly allowance, no top-up fee. BYOK stays free.

free app · your subscriptions · local-first