What does OpenRouter charge on top of model prices?

How credit purchase fees and pass-through pricing work

When you call a model through OpenRouter, the prompt and completion rates match the underlying provider's standard pricing. OpenRouter passes through those inference prices without any markup on every plan, so a token run through Anthropic or OpenAI costs the same base rate on OpenRouter as it does on their own dashboards.

You pay OpenRouter's platform cut when you load money onto your balance. On the Standard plan, OpenRouter adds a 5.5% platform fee to your credit purchase. If you deposit $100, you pay $105.50 and receive $100 in usable credits. On the Business plan, that platform fee rises to 8%. The fee covers the platform layer, routing, unified keys, and provider fallbacks.

Total charge = Purchased credits + (Purchased credits * Platform fee percentage)

Because the fee applies to the deposit rather than the API request, every individual model call simply deducts the provider's exact token price from your credit balance. OpenRouter does not offer volume discounts on inference costs, though Enterprise accounts can negotiate fee discounts on the OpenRouter platform fee itself.

There is one way to lower your token bill: OpenRouter has an opt-in setting that logs your prompts and completions in exchange for a 1% discount on usage costs. If your workload permits public or platform logging, you shave 1% off the pass-through model price.

All pricing is denominated in US dollars. Depending on the model architecture, rates are quoted per million prompt tokens, per million completion tokens, per request, per image, or for reasoning tokens. Streaming responses cost the exact same per-token rate as standard non-streaming calls.

Taxes are handled separately. OpenRouter prices are exclusive of applicable taxes, and VAT or GST may be added to your invoice where required by local tax law.

OpenRouter plan tiers compared

OpenRouter offers four account tiers: Free, Standard, Business, and Enterprise. Standard and Business operate as pay-as-you-go setups with no minimum monthly spend and no long-term lock-in.

Upgrading to Business is self-serve in your account settings with no sales calls or contracts. You can downgrade back to Standard 30 days after upgrading.

For payment methods, Standard and Business accounts accept major credit and debit cards, AliPay, and cryptocurrency payments in USDC. Enterprise plans support invoicing and purchase orders.

Unused credits are subject to strict terms. OpenRouter reserves the right to expire unused credits after one year of purchase. If you top up by mistake, you can request a refund for unused credits within 24 hours of the transaction by using the refund button on the Credits page. Platform fees are non-refundable, and cryptocurrency payments in USDC are never refundable. Deleting your OpenRouter account via Settings > Manage Account > Security permanently wipes any unused balance.

Bring Your Own Key (BYOK) fees and limits

If you already have direct enterprise contracts or existing credits with providers like OpenAI or Anthropic, you can plug your own API keys into OpenRouter.

With BYOK, the underlying provider bills your account directly for inference tokens. OpenRouter does not sit in the billing chain for that model spend.

BYOK is not supported on the Free plan, but Standard and Business plans both come with a monthly allowance of $25,000 in list-price inference. You run queries against your own keys through OpenRouter's router, and OpenRouter charges zero platform fee on that initial $25,000 slice every month.

Once your BYOK usage crosses $25,000 in list-price inference within a calendar month, OpenRouter bills a 5% fee based on what the request would have cost at list price on OpenRouter. This 5% surcharge is deducted from your prepaid OpenRouter credit balance, so you need to maintain credits in your OpenRouter wallet even when routing to your own provider keys. On Enterprise plans, BYOK allowances and overage rates are customized by contract.

Free models, request caps, and the 402 error

The Free plan provides access to 25+ free models supported across 4 free providers. Paid plans unlock the full catalog of 600+ models across 80+ providers.

Free models are identified in the catalog with the :free suffix, such as specific open-weight variants, or you can route to them dynamically using the Free Models Router slug openrouter/free.

Using free models comes with two tiers of rate limits:

• Under $10 in credit purchases: Capped at 20 requests per minute and 50 requests per day.
• $10 or more in credit purchases: The daily ceiling expands to 1,000 requests per day.

To absorb rounding differences and top-up processing fees, OpenRouter grants the 1,000 requests per day ceiling starting one credit below the $10 threshold.

Keep in mind that failed attempts count toward your daily cap on free models. Furthermore, if your prepaid credit balance drops below zero from paid requests, OpenRouter blocks access to free models and returns 402 errors until you deposit credits to restore a positive balance.

For paid models, OpenRouter imposes an in-flight spending budget on newer accounts without established spending history or accounts with balances below an internal threshold. OpenRouter calculates estimated token costs (prompt tokens plus max_tokens) before dispatching a request; if that estimate exceeds your available in-flight budget alongside currently running requests, the call is rejected with a 402 status code before reaching the provider.

This in-flight budget hold looks only at raw input and completion token rates. Per-request fees, plugin charges, and image generation costs are excluded from the in-flight estimate. Free models and pure BYOK requests without paid plugins are exempt from in-flight budget holds.

You can inspect your API key's current limits, credit balances, and free model usage by querying the key endpoint:

curl -fsS https://openrouter.ai/api/v1/key \
  -H "Authorization: Bearer $OPENROUTER_API_KEY"

If you manage API keys programmatically, management key operations (POST /api/v1/keys, PATCH /api/v1/keys/{hash}, and DELETE /api/v1/keys/{hash}) share a pooled limit of 10,000 requests per day, resetting at 00:00 UTC. Budget guardrails also operate on UTC windows: daily budgets reset at 00:00 UTC, weekly budgets reset Mondays at 00:00 UTC, and monthly budgets roll over on the first day of the month at 00:00 UTC.

Routing variants and failure billing

OpenRouter allows you to append routing tags to model strings to customize how traffic flows:

• :floor sorts available providers by lowest price first.
• :nitro sorts providers by throughput.
• :exacto sorts providers using quality-first signals optimized for tool-calling reliability.
• :batch targets the batch-priced endpoint used by the Batch API.

Using the general router openrouter/auto adds no extra fee; you pay the standard token rate of whichever model fulfills the request.

If a provider errors out during a call and OpenRouter falls back to another provider, you are not charged for the failed run. You only pay for the model run that answers.

Under OpenRouter's Zero Completion Insurance, if a model returns an empty body and no finish reason, OpenRouter voids the token charge for that response. However, auxiliary services that executed before the failure - like web search plugins or file parsing - may still be billed to your credits.

To inspect raw, un-marked-up pricing across every catalog entry at any time, you can pull the model directory directly without an API key:

curl -fsS "https://openrouter.ai/api/v1/models?output_modalities=all"

Monitoring OpenRouter spend with Costbrake

If you run OpenRouter alongside other infrastructure, tracking usage manually across dashboards gets messy quickly.

Costbrake watches spending across 23 cloud, AI, and SaaS platforms, including OpenRouter, OpenAI, Anthropic, AWS, and Google Cloud. Checks run hourly on the Free plan, every 5 minutes on Pro and Team, and every minute during detected spikes. Costbrake triggers spike alerts when an hour exceeds 3x your 7-day baseline and costs at least $5 more than usual, alongside budget alerts at 50%, 80%, and 100%.

For OpenRouter, Costbrake reads key counters, remaining credits, and workspace activity using either a read-only management key or a single API key. If a runaway loop kicks off, Costbrake offers two opt-in brakes for OpenRouter that an admin can trigger:

1. Disable API key: Instantly halts every request made with that key while leaving other keys running.
2. Set credit limit to $0: Blocks paid model requests while allowing free models to continue functioning.

You can set up monitoring via your terminal with agent setup or configure key-level limits manually by following our guide on how to set a credit limit on an OpenRouter API key. If you deploy multiple providers directly, see how to monitor spend across OpenAI, Anthropic, Gemini and other AI APIs in one place. Paid plans start at a flat $12 a month with no cut of your spend; see Costbrake pricing for details.