OpenRouter Pricing 2026: Fees, Credits & Top-Ups

OpenRouter does not mark up model prices — but it is not free either. The platform makes money on credit top-ups (5.5%, minimum $0.80) and on BYOK usage past the free allowance (5%), while inference itself is billed at the upstream provider's list price ^[1].

The whole fee model in five lines:

• Inference: exact provider list price per token, no markup ^[1]

• Buying credits: 5.5% fee, $0.80 minimum (Stripe); about 5% when paying with crypto

• BYOK (your own provider keys): first 1M requests/month free, then 5% of what the call would have cost on OpenRouter

• Free models: 50 requests/day under $10 lifetime credits, 1,000/day once you have bought $10+ ^[1]

• Routing: defaults balance price and uptime; price-floor routing (:floor) or max_price caps can push each request to the cheapest provider ^[2]

Below we break each piece down with worked examples, then compare the total against a discount-based aggregator.

Key Takeaways

• OpenRouter bills inference at the upstream provider's list price with no markup and earns revenue from credit top-up fees and from BYOK usage beyond the free monthly allowance.
• Stripe credit top-ups cost 5.5% with a $0.80 minimum, so a $10 top-up costs $10.80, while crypto via Coinbase runs about 5%; model effective token price as list price × 1.055.
• With BYOK, the first 1M requests per month are free, after which OpenRouter deducts 5% of the equivalent platform cost from your credit balance.
• Free models allow 50 requests/day across all free models for accounts under $10 in lifetime credit purchases, rising to 1,000/day after a one-time $10 purchase, with no uptime guarantees.
• Use :floor or provider: {"sort": "price"} for cheapest-first routing and max_price for hard caps; in the worked example, a $325.00 GPT-5.2-class workload totals about $342.88 on OpenRouter.

Related reading: OpenRouter API Guide: One API for Top AI Models; OpenRouter vs Direct Model APIs: Which Is Better?

The three ways OpenRouter charges you

Pass-through inference pricing

Every model page lists input/output prices per million tokens, and those numbers match the upstream provider's own price sheet ^[3]. If GPT-5.2 costs $1.25 in / $10 out per 1M tokens at OpenAI, that is what OpenRouter deducts from your credits — the aggregation layer itself adds zero to the per-token rate.

The credit top-up fee

Credits are prepaid. Each Stripe top-up costs 5.5% of the amount, with a $0.80 floor — so a $10 top-up costs $10.80 (an effective 8%), while a $500 top-up costs $527.50. Crypto payments via Coinbase run about 5%. This is the fee most cost comparisons forget: your effective token price is list price × 1.055.

BYOK surcharges

Plugging in your own OpenAI or Anthropic key lets you use provider credits and negotiated rates. The first 1M BYOK requests per month are free; beyond that OpenRouter deducts 5% of the equivalent platform cost from your credit balance ^[1]. For most small teams the allowance is effectively unlimited; at serious scale the 5% becomes a line item.

Free tier and rate limits

The 50-per-day baseline

Models tagged :free cost nothing but are throttled hard: accounts with less than $10 of lifetime credit purchases get 50 free-model requests per day across all free models combined.

Unlocking 1,000 requests per day

A one-time purchase of $10 in credits raises the ceiling to 1,000 free-model requests/day ^[1] — the cheapest way to make the free tier useful for prototyping.

What free actually buys

Free variants are typically lower-priority capacity with tighter context limits and no uptime guarantees. Treat them as a sandbox, not a production tier.

How routing decisions change your bill

Default load balancing

By default OpenRouter spreads requests across providers weighted toward cheaper, healthier endpoints, so two identical requests can be served by different upstreams at slightly different prices ^[2].

Price-floor routing

Appending :floor to a model slug (or provider: {"sort": "price"} in the request body) forces cheapest-first routing — useful for batch workloads where latency matters less than cost.

Hard price caps

The max_price field rejects any provider quoting above your limit, e.g. {"max_price": {"prompt": 1, "completion": 8}} (dollars per 1M tokens). It is the only way to make "never pay more than X" a guarantee rather than a preference.

Worked example: what a workload really costs

Numbers use $1.25/1M input and $10/1M output list pricing; APIMart applies its ~20% platform discount — check live per-model rates on the APIMart pricing page.

When the 5.5% is worth it

If OpenRouter saves you from maintaining three vendor integrations and gives you failover for free, $18 on a $325 bill is cheap insurance. The fee only starts to hurt at five-figure monthly spend.

When it is not

Single-model, high-volume workloads gain nothing from aggregation fees — go direct, or use a gateway whose unit prices are actually below list.

The discount-aggregator alternative

APIMart flips the model: instead of list price plus a top-up fee, models are listed at roughly 20% below official prices, across chat, image, video and audio. The pricing page shows both the official and discounted rate per model, so the comparison is one click, not a spreadsheet.

Same Models, ~20% Below Official Pricing

APIMart aggregates 500+ chat, image, video and audio models behind one OpenAI-compatible API — with per-model prices listed transparently below official rates.

Compare Model Prices

Takeaways

OpenRouter's pricing is genuinely transparent: list-price inference, a visible 5.5% top-up fee, and a 5% BYOK surcharge past 1M monthly requests ^[1]. Model your real cost as list price × 1.055, use :floor and max_price to keep routing honest, and if your stack extends beyond text models — or the fee math stops working for you — a discount-based multimodal gateway like APIMart is the natural next comparison.