Fazlur Rahman
Quick Answer
OpenRouter pricing 2026: pay-as-you-go 5.5% credit-purchase fee with a $0.80 minimum per top-up; BYOK (Bring Your Own Key) free up to $25,000 of monthly list-price inference, then 5%; Enterprise free up to $200,000/mo; 300+ models from 80+ providers. Token rates pass through with no separate per-token markup.
Last verified: Sep 16, 2026.
At a glance
- Shared credit fee: 5.5% on credit purchases (5% crypto), $0.80 minimum
- BYOK: free up to $25,000/mo list-price inference, then 5%
- Enterprise: free up to $200,000/mo, then negotiated
- Per-token markup: none — provider rates pass through
- Free tier: 25+ models, 20 req/min, 50-1000/day by credit balance
- Real cost premium: 5-7% above direct billing
- Failed-request policy: no charge on fallback
- Catalogue: 300+ models, 80+ providers on Pay-as-you-go
Two fee models, not one
OpenRouter's pricing has two independent fee models — shared credits and BYOK — and the choice between them determines whether you pay 5.5% or 0%.
Shared credits work like a prepaid phone: you buy $100 of credits, OpenRouter takes 5.5% ($5.50) immediately, and the remaining $94.50 is spent on inference at the provider's per-token rate. Every subsequent request draws down the credit balance at the provider's rate with no per-call fee. The 5.5% is charged only when you buy credits, not on individual API calls. The $0.80 minimum per purchase means a $5 top-up actually leaves $4.20 for inference — a 16% effective fee instead of 5.5% (OpenRouter pricing, September 2026).
BYOK routes API calls through your own provider keys (OpenAI, Anthropic, Google). OpenRouter charges no fee on the first $25,000 of monthly list-price inference — your bill is the provider's direct rate. Above $25,000, OpenRouter adds 5% of what the same call would have cost on its platform. Enterprise raises the threshold to $200,000. BYOK is the cheapest route for any team spending above ~$15/month that already has provider accounts (ofox.ai, August 2026).
The 5.5% routing tax in practice
Across four representative workloads, OpenRouter's shared-credit 5.5% fee adds a constant percentage tax that does not change with volume — and the $0.80 minimum makes it regressive for small bills.
For a heavy code agent on Claude Sonnet 4.6 at $186/month direct, OpenRouter bills $196.23 — a $10.23 routing tax (5.5%). For an RAG workload on GPT-5.4-mini at $62.70/month direct, OpenRouter bills $66.15 — a $3.45 tax (5.5%). For a Mistral Medium 3.5 structured-output workload at $60/month, OpenRouter bills $63.30 — a $3.30 tax (5.5%). Across the four-workload network total of $312.66 direct, OpenRouter adds $17.78 in routing tax (5.7%) (BudgetForge, July 2026).
For a low-volume Haiku 4.5 chatbot at $3.96/month direct, OpenRouter bills $4.76 — a $0.80 minimum tax (20.2%). The minimum dominates small bills. To reduce the effective rate, buy credits in bigger, less frequent chunks. The 1M free requests per month on Pay-as-you-go also helps — for low-volume apps with high request counts, the free tier absorbs most of the bill (Markaicode, August 2026).
Free failover is the strongest argument for shared credits
The 5.5% credit fee buys something that direct provider access cannot match: free failover across providers.
OpenRouter charges nothing for failed or fallback requests when routing is enabled. If your primary model fails (rate limit, server error, timeout), OpenRouter falls back to a backup provider and bills only the successful call. For an agent that retries 3 times across providers, a 30% primary failure rate means 30% of inference is free — often worth more than the 5.5% credit fee in raw dollars (ofox.ai, August 2026).
The failover story is the single strongest case for paying the 5.5%. For multi-model apps that need cross-provider fallback for reliability, the operational savings outweigh the routing tax. For a single-model high-volume app, direct provider access is cheaper and simpler (BudgetForge, July 2026).
BYOK vs shared credits: the crossover math
BYOK's fee structure ($0 up to $25,000/mo, then 5%) is never higher than shared credits' flat 5.5% at any volume from $0 to $50,000 — so BYOK is fee-wise equal-or-cheaper throughout.
The real BYOK trade-off is operational. BYOK requires managing separate provider keys, quotas, rate limits, and billing for each model you use. For a team that needs unified billing and operational simplicity, shared credits at 5.5% is worth it. For a team that already has provider accounts and can manage multiple keys, BYOK removes the platform fee entirely (Markaicode, August 2026).
Default to BYOK once monthly spend is predictable and above a few hundred dollars — it removes the platform fee entirely below $25,000/month. Reserve shared-credit pay-as-you-go for prototyping, low-volume side projects, or teams that would rather pay 5.5% than manage separate provider keys (Markaicode, August 2026).
What enterprise buyers should do next
Three actions for organisations evaluating OpenRouter in 2026.
- Decide BYOK vs shared credits by volume. Below $15/month or for one-off prototyping, shared credits. Above $15/month with predictable spend, BYOK. Above $25,000/month, direct provider billing on the dominant model — the BYOK overage kicks in.
- Use the model catalogue as a routing layer. OpenRouter's real value is multi-model access with one integration. Use it for fallback routing, A/B testing across providers, and unified logging. Do not use it as a price-discovery tool — provider rates are public and OpenRouter passes them through.
- Buy credits in bulk. The $0.80 minimum on every credit purchase makes small frequent top-ups expensive. Buy $500 chunks once a month instead of $50 chunks ten times. The effective rate drops from 5.5% nominal to closer to 5.0% amortised.
What to watch next
Three near-term datapoints. First, OpenRouter's negotiated Enterprise pricing above $200,000/mo — current published rates only go up to the free threshold, with anything above custom-quoted. Second, the introduction of OpenRouter-published SLA tiers — currently OpenRouter has no standing uptime SLA on shared credits; Enterprise adds contractual SLAs. Third, OpenRouter's continued expansion of open-weight model coverage as new Llama, Mistral, and Qwen families ship — every new open-weight model typically appears on OpenRouter within 24-48 hours of release.
