Summary: OpenRouter does not mark up model prices; you pay each provider's list rate per token, and OpenRouter takes its cut as a fee when you buy credits (5.5% on the Standard pay-as-you-go plan, 8% on Business). The free plan gives you 25+ free models at 50 requests per day, which jumps to 1,000 per day once you have bought at least 10 credits in total. If you already have provider keys, bring-your-own-key (BYOK) usage is fee-free up to $25,000 of list-price inference a month.
I went through OpenRouter's official pricing page, its limits and BYOK documentation, the FAQ, and the public models API on the day this was written. Below is what each plan actually costs, where the free tier stops being useful, and the billing errors people run into most.
What OpenRouter is, in one paragraph
OpenRouter is a single API (OpenAI-compatible, with the base URL https://openrouter.ai/api/v1) that sits in front of hundreds of models from dozens of providers. You load credits once, use one API key, and switch models by changing a model ID string such as openai/gpt-4o-mini. If a provider errors or is rate limiting you, OpenRouter can fall back to another provider serving the same model. The pricing question is therefore not "how much is OpenRouter" but "what do I pay on top of the model's own price, and what limits apply."
The four plans at a glance
The pricing page lists Free, Standard (pay-as-you-go), Business (pay-as-you-go) and Enterprise. There is no monthly subscription on the first three. Standard and Business have no minimum spend and no lock-in, according to the pricing FAQ.
| Item | Free | Standard (pay-as-you-go) | Business (pay-as-you-go) |
|---|---|---|---|
| Platform fee | N/A | 5.5% on credit purchases | 8% on credit purchases |
| Models | 25+ free models | 600+ models | 600+ models |
| Providers | 4 free providers | 80+ providers | 80+ providers |
| Rate limits | 50 requests/day | Passthrough from the provider(s) | Passthrough from the provider(s) |
| BYOK | No | $25,000 list-price inference/month with no fee, 5% after | Same as Standard |
| Budgets and spend controls | No | Yes | Yes |
| Prompt caching | No | Yes | Yes |
| EU / US in-region routing | No | No | Yes |
| Workspace limit | 5 | 5 | 1,000 |
| Support | Community |
Source: openrouter.ai/pricing, checked 2026-11-10. Enterprise is contract-priced with fee discounts, SSO/SAML, contractual SLAs and invoicing.
How the money actually flows
1. Token prices are the provider's list price
The pricing FAQ is explicit: inference is billed at the provider's list price on every plan, and OpenRouter does not add a per-request markup. Each model page shows a price per million input tokens and per million output tokens; some models also charge per request, per image, or for reasoning tokens. Streaming does not change the price.
2. The fee is charged when you top up
OpenRouter's revenue on Standard and Business is the platform fee on credit purchases: 5.5% on Standard and 8% on Business. As a quick piece of arithmetic, a $20 top-up on Standard carries a $1.10 fee (5.5% of $20), and the same top-up on Business carries $1.60. The FAQ says the fee is "charged when you buy credits, not on individual API requests," so every model costs the same per token on both plans. The only reasons to pay the higher Business fee are the features in the table, such as in-region routing, workload identity federation, advanced notifications, and the 1,000-workspace limit.
3. A worked cost example
Using the price the public models API listed for openai/gpt-4o-mini on 2026-11-10 ($0.15 per million input tokens and $0.60 per million output tokens), here is what a small batch job would cost. This is plain arithmetic, not a benchmark I ran:
- 1,000 requests, each with 2,000 input tokens and 500 output tokens.
- Input: 2,000,000 tokens at $0.15/M = $0.30.
- Output: 500,000 tokens at $0.60/M = $0.30.
- Total inference: about $0.60, deducted from your credit balance.
The platform fee was already paid when you bought the credits, so it does not show up per request. Prices change, so always re-check the model page before you budget.
4. Failed and fallback attempts
According to the pricing FAQ, an attempt that errors is not charged for its model tokens, so when fallback kicks in you pay only for the run that answers. Auxiliary services that already ran, such as web search or file parsing, may still be billed. Routing through openrouter/auto adds no extra fee; you pay the rate of whichever model serves the request.
The free tier: what you really get
Free model variants have IDs ending in :free. The limits documentation sets these caps:
- Fewer than 10 credits purchased (all time): 20 requests per minute and 50 requests per day.
- At least 10 credits purchased (all time): 20 requests per minute and 1,000 requests per day.
The tier is chosen by credits you have ever purchased, not your current balance. That makes a one-time $10 top-up the cheapest upgrade OpenRouter has: you raise your free-model ceiling twentyfold and still have the credits to spend on paid models later.
When I queried the public models API on 2026-11-10, it listed 13 chat or multimodal model IDs ending in :free (including Google's Gemma 4 models and several NVIDIA Nemotron models), plus an openrouter/free router that picks a free model at random. The pricing page's "25+ free models" count also includes free embedding, rerank, speech and other model types shown on the models page. The free list moves quickly; one free speech model on the models page was already labelled as going away on November 30, 2026, so don't build anything important on a single free ID.
OpenRouter's own FAQ describes the free models as having low rate limits and being "usually not suitable for production use." Treat them as a testing playground.
BYOK: when bringing your own keys pays off
With bring-your-own-key, you add your OpenAI, Anthropic, Google Vertex, Bedrock or other provider key in your workspace BYOK settings. The provider bills you directly for inference, and OpenRouter charges 5% of what the same request would cost at list price on OpenRouter, deducted from your OpenRouter credits. On Standard and Business, the first $25,000 of list-price inference each month through BYOK carries no OpenRouter fee at all.
By default, OpenRouter tries your key first and falls back to its shared capacity if your key fails. You can also mark a key so that shared capacity is never used, which matters if you have data-handling agreements with a provider. For most individuals and small teams, BYOK under the $25,000 allowance effectively makes OpenRouter a free routing and logging layer on top of accounts you already pay for.
How to start without surprises
- Create an account and an API key, then call a
:freemodel first to confirm your code works. - Check your limits with
GET https://openrouter.ai/api/v1/key. The response includeslimit_remaining, usage counters, and afree_model_daily_requestsobject with used, limit and remaining. - Set a per-key credit limit before you share a key with a teammate, an agent, or a CI job.
- If you plan to use free models daily, buy 10 credits once to unlock the 1,000-requests-per-day tier.
- If you already pay a provider directly, add that key under BYOK rather than buying extra credits for the same model.
Editor's note
Good fit: developers who want to compare many models behind one OpenAI-style API, side projects that need free models for testing, and teams that want spend caps per key without managing five provider dashboards.
Not a great fit: production apps that rely only on free models (50 or 1,000 requests a day is a hard ceiling), or companies that need SSO/SAML and contractual SLAs but don't want an Enterprise contract.
Gotchas I found in the docs:
- 402 even on free models: a negative account balance can block free models too. Top up to bring the balance above zero.
- 402 with a positive balance: newer or low-balance accounts have an "in-flight spending budget." If many paid requests run at once, you can get
in_flight_budget_exhausted; wait for theRetry-Afterheader. If the error saysweight_exceeds_budget, lowermax_tokensor the prompt size, because retrying will not help. - 403: a budget set by an org admin or on the key itself has been spent.
- 429: either OpenRouter's free-model cap or the upstream provider. Back off and retry, add fallback models, or move to the paid variant, which has no platform-level request cap.
- More keys won't help: the docs say extra accounts or API keys do not raise rate limits, because capacity is governed globally.
- Refunds: unused credits can be refunded only within 24 hours of purchase, platform fees are non-refundable, and crypto payments are never refundable. OpenRouter also reserves the right to expire unused credits one year after purchase.
FAQ
Is OpenRouter cheaper than calling the provider directly?
Per token it is the same, because OpenRouter passes through list prices. You pay the 5.5% fee on credit purchases on Standard, so for a single model you already use heavily, going direct (or using BYOK) is usually slightly cheaper. The value is in switching models, fallbacks and unified billing.
Do I need a credit card for the free models?
No purchase is needed to use :free models at 50 requests per day. Buying at least 10 credits once raises that to 1,000 per day.
Are taxes included in the prices?
No. The pricing FAQ says prices exclude applicable taxes, and VAT or GST may be added on invoices where required.
Related reading on this blog: LiteLLM Tutorial: One Proxy for OpenAI and Ollama if you'd rather self-host your router, Together AI Pricing: Credits, Rates and Limits, and Groq API Free Tier: Rate Limits, Costs and When to Upgrade.
Start on free models, buy 10 credits once when the 50-a-day cap bites, and switch to BYOK when you already pay a provider.
See the official screenshots at the end of this post. Image credit: screenshots of openrouter.ai pricing, documentation and models pages, captured 2026-11-10.
Last verified: 2026-11-10
Sources: OpenRouter pricing, Credit limits and rate limits, BYOK documentation, OpenRouter FAQ, Free models list, Quickstart.
댓글
댓글 쓰기