AI cost · OpenRouter
OpenRouter cost: what you actually pay
OpenRouter passes through the underlying model's own price with no inference markup, so the real cost creep is the credit top-up fee (5.5%/5%), silent routing to a different-priced provider variant of 'the same' model, and BYOK fees past the free monthly quota.
Public pricing
Token rates (USD per 1M tokens)
| Model | Input | Output | Cache read |
|---|---|---|---|
| openai/gpt-5passthrough of OpenAI's own listed rate | $1.25 | $10 | — |
| deepseek/deepseek-chat (DeepSeek V3)passthrough rate for the routed provider at fetch time; varies by which backing host serves the request | $0.2288 | $0.9144 | — |
OpenRouter is a router/marketplace, not a model host: per-token prices equal the underlying provider's own rate ('no markup on inference pricing' per OpenRouter's own FAQ). OpenRouter instead earns via a credit-purchase fee (5.5% Stripe / 5% crypto) and, for BYOK (bring-your-own-key) usage, a 5% fee on usage beyond the first 1,000,000 free requests/month. The same model name can route to multiple provider backends (different quantizations/hosts) at different prices unless a specific provider is pinned.
Last updated: 2026-07-22. Sourced from OpenRouter Docs/FAQ (official) ↗, GPT-5 on OpenRouter (official) ↗.
Where the real OpenRouter cost hides
The sticker is the smallest part of the story. For OpenRouter, these are the line items that quietly inflate the bill:
- Prepaying credits via Stripe costs an extra 5.5% on top of whatever the model itself costs — an easy-to-miss 'top-up tax' when comparing OpenRouter's sticker price to going direct to the model vendor.
- The same model ID can route to different underlying providers/quantizations (e.g., different DeepSeek hosts) with different prices and quality; the effective price can shift request to request unless a provider is pinned.
- Free (':free' suffix) model variants carry very small rate limits and commonly permit upstream logging/training use of prompts — fine for prototyping, a poor fit for production traffic that then migrates to the paid variant at full price.
- BYOK looks free but only for the first 1M requests/month; heavy BYOK users pay a 5% OpenRouter fee on top of already paying the underlying provider directly.
What an audit finds on OpenRouter
Teams price a workload off OpenRouter's listed per-model rate but forget the credit top-up fee and that default/'auto' routing can silently substitute a pricier or lower-quality provider variant of the same model during high demand.
Worked example: OpenRouter at production volume
Take 50M input and 10M output tokens a month — a modest production workload, and a round number we picked — on openai/gpt-5 at the published rates above.
- Monthly
- $163
- Annual
- $1,950
- On deepseek/deepseek-chat (DeepSeek V3)
- $21
Same volume, $1,703 a year apart. Most workloads are a mix — some of it genuinely needs the frontier model and some of it does not — so the real question is what share of your calls is the routine kind, and that is a measurement, not a guess.
Arithmetic on the published prices above. Your mix will differ — run your own numbers.
OpenRouter pricing FAQ
Does OpenRouter mark up the model price?
No stated markup on the inference rate itself — OpenRouter's docs say pricing passes through from the provider; OpenRouter monetizes via the credit-purchase fee and a small BYOK usage fee.
Is OpenRouter cheaper than going direct to the model vendor?
Rarely — for a given model it's typically the same $/token as the provider's direct API, plus OpenRouter's credit top-up fee, so it's an aggregation/convenience layer rather than a discount.
How much can routing cut a OpenRouter bill?
On 50M input and 10M output tokens a month, openai/gpt-5 costs $163 at published rates and deepseek/deepseek-chat (DeepSeek V3) costs $21 — $1,703 a year apart for the same volume. What you can actually move depends on how much of your traffic is routine, which is a measurement rather than a guess.
Are these OpenRouter prices current?
They were last checked against OpenRouter Docs/FAQ (official) and GPT-5 on OpenRouter (official) on 2026-07-22. Every price on this page carries that date and a link to its source, and a scheduled check fails our build when any figure goes more than 30 days unverified. Vendors do change pricing between checks — the source link is there so you can confirm before you sign.
What is OpenRouter really costing you?
Spendassay connects your OpenRouter spend to actual usage and engineering output — a proof-level AI cost report with the wasted dollars named. Neutral across every AI vendor.
Free · read-only · no card
Other api platforms pricing
More: the complete AI cost management guide · every tool, with prices · cost per outcome · compare platforms · how we measure