OpenRouter
OpenRouter pricing works differently from everything else in this index.
Updated on:
OpenRouter pricing: a gateway that doesn't take a cut of the tokens
OpenRouter pricing works differently from everything else in this index. It resells other companies' models and takes nothing from the tokens: rates match the provider's list price exactly. It runs on a platform fee charged when you buy credits: 5.5% with a $0.80 minimum on Standard, 8% on Business and 5% flat on crypto. The fee is the spread, and it's published.
Key takeaways
OpenRouter charges no markup on inference. Claude Sonnet 5 bills $2 per million input and $10 output, identical to Anthropic's own rate.
The platform fee lands on the credit purchase, not the request: 5.5% with a $0.80 minimum on Standard, 8% on Business, 5% flat on crypto.
Bringing your own provider key costs 5% of what the same model and provider would normally cost on OpenRouter, above a $25,000 monthly allowance.
The same model costs different amounts per provider, and default routing picks by price weighted against recent outages.
OpenRouter pricing in 2026
Plan | Platform fee | Workspaces | BYOK allowance | Rate limits
|
|---|---|---|---|---|
Free | N/A | 5 | N/A | 50 requests/day, 20 rpm, free models |
Standard | 5.5% of the purchase, $0.80 minimum | 5 | $25,000 of list-price inference a month, 5% after | None on paid models |
Business | 8% of the purchase | 1,000 | $25,000 a month, 5% after | None on paid models |
Enterprise | Custom, fee discounts available | Custom | $200,000 a month, 5% after | None, plus your own capacity |
What OpenRouter actually meters
OpenRouter meters tokens at the upstream provider's rate and takes its cut somewhere else. The FAQ says it plainly: "We do not mark up provider pricing." We checked. On 22 September 2026 the public models API returned Claude Haiku 4.5 at $1 and $5 per million, Sonnet 5 at $2 and $10, Opus 5 at $5 and $25, Fable 5.1 at $10 and $50. Every figure matches Anthropic's card to the cent.
What varies is which provider serves you. DeepSeek V4.1 Flash listed 23 endpoints running $0.12 to $0.375 per million input tokens and $0.42 to $1.50 output, a 3.1x spread on input and 3.6x on output. The default strategy weights each candidate by the inverse square of its price after dropping anyone with an outage in the last 30 seconds. Override it with sort, max_price or :floor.
Tokens aren't the only meter. Server tools bill per request, from $1 per 1,000 on the fastest search modes to $15 per 1,000 on deep reasoning. A batch submitted to the Batch API generally bills at 50% of the provider's per-token price.
How credits work
Credits are the only currency, in US dollars, and the fee attaches to the purchase, not usage. Buy $100 of credits on Standard and you pay $105.50, or $108 on Business, for identically priced tokens. The $0.80 minimum makes small top-ups expensive: a $5 purchase pays 16%.
Expiry is the sharp edge. OpenRouter's terms reserve the right to expire unused credits one year after purchase. Refunds run on a 24-hour window, platform fees never come back, and crypto payments aren't refundable. Auto top-up refills the balance below a threshold you set.
Two mechanics work in your favour. Zero Completion Insurance means a response with no output tokens costs nothing, even when the provider bills OpenRouter for prompt processing. Failed and fallback attempts aren't billed, so a request that tries three providers pays for one.
What happens when you hit the limit
Nothing auto-charges. Credits hit zero, the API returns 402 Payment Required, and requests stop until you top up. That's a hard stop, which is what prepaid should mean.
Rate limits are separate and tied to the balance. Without credits you get 50 requests a day and 20 a minute on free models. Buy at least $10 and that rises to 1,000 a day. Paid models carry no OpenRouter rate limit, though the provider's own limits apply. Workspace budgets cap spend per interval, with a flag for whether BYOK counts.
How OpenRouter pricing has changed across all these years
Date | Milestone | Source
|
|---|---|---|
22 Sep 2026 | Batch API ships on 70+ models. Providers generally charge 50% of their per-token price for a 24-hour turnaround window | Vendor |
Aug 2026 | BYOK allowance switches from request count to list-price inference cost: $25,000 a month on Standard, $200,000 on Enterprise | Vendor |
30 Apr 2026 | Response caching ships. Identical repeat requests bill zero tokens | Vendor |
1 Oct 2025 | BYOK gets 1,000,000 free requests a month, 5% beyond | Vendor |
9 Jun 2025 | Fixed $0.35 Stripe fee removed. Fee becomes 5.5% with a $0.80 minimum, or 5% flat on crypto | Vendor |
25 Mar 2025 | Zero Completion Insurance extends to every model. Responses with no output tokens stop being charged | Vendor |
20 Dec 2024 | BYOK launches at 5% of the upstream provider's cost | Vendor |
Customer
Sentiment Highlights
"I like the product, and I think the fee is fair, but if you want the absolute best pricing then go direct."
cedws, Hacker News, April 2026
"I'm probably overpaying by more than the 5.5% fee"
wongarsu, on limiting providers, Hacker News, April 2026
Explore other providers
ChatGPT
Chatbot
ChatGPT pricing sells seats from $0 to $100 a month and meters the ambitious work on top in credits.
OpenAI API
API
OpenAI API pricing charges per million tokens, separately for input and output, with a different rate for every model.
Claude
Chatbot
Claude pricing sells seats from $0 to $200 a month, and each buys an allowance Anthropic describes only as a multiple of the tier below.
Does OpenRouter mark up model pricing?
How much is OpenRouter's platform fee?
Does OpenRouter charge for BYOK?
Do OpenRouter credits expire?
