OpenRouter gives you one API key and one OpenAI-style endpoint for 500+ models from 80+ providers, with automatic fallback when a provider fails. Token prices match the providers' list prices; OpenRouter charges a 5.5% fee when you buy credits. Pick it to try many models from one app. The price is a second company in your data path.
Other good fits
Keeping an app running through a provider outage: OpenRouter falls back to another host of the same model by default.
Capping spend per API key, for example one key per colleague or per project.
+ Pros
Pick it when you want to compare models on your own task. You change one model value and keep the rest of your code.
Pick it when an outage at one provider shouldn't stop your app. Fallback to other providers of the same model is on by default.
Pick it when you want one usage log and one bill for every model, with a spending limit on each key.
Pick it when you want a free prototype. There are 25+ free models, limited to 50 requests a day, or 1,000 a day once you've bought at least $10 of credits.
− Cons
Read cons
Pick something else when you only use Claude. The Claude API directly gives you Anthropic's own features and one less company in the path.
Pick something else when data must stay in the EU on a small budget. Requesty includes an EU endpoint on every plan, and Scaleway runs its models in Paris.
Pick something else when you want to run the gateway yourself with no fee. LiteLLM is open source.
One caution: OpenRouter doesn't log prompts by default, but the provider that serves a request has its own data policy. For sensitive work, turn on data-policy or zero-retention routing so only matching providers are used.
Avoid if
Personal or confidential data must be processed in the EU and you're on the Standard plan. EU in-region routing needs the Business plan (8% fee).
You need a fixed monthly price. Billing is prepaid credits spent per token.
Connection
Send JSON to https://openrouter.ai/api/v1/chat/completions with Authorization: Bearer $OPENROUTER_API_KEY. Use a current model slug; meta-llama/llama-3.3-70b-instruct is documented. OpenAI-compatible does not mean identical features: use provider.require_parameters: true when unsupported parameters must exclude a route. Current model availability was not tested.
Billing
Read billing
Prices checked 23 September 2026 (openrouter.ai/pricing):
Plan
OpenRouter fee
What you get
Free
none
25+ free models, 50 requests a day
Standard (pay as you go)
5.5% on credit purchases (minimum $0.80); 5% when paying in crypto
500+ models, budgets and spend controls, data-policy routing
Business (pay as you go)
8% on credit purchases
adds EU and US in-region routing, up to 1,000 workspaces
Enterprise
by contract
SSO, SLAs, a custom DPA
Model tokens cost the provider's list price on every plan, with no markup. If you bring your own provider keys, the first $25,000 of list-price use a month carries no fee, then 5%.
Running effort is small. Someone tops up credits (or turns on auto top-up), sets a limit on each key and checks the Activity page. The bill grows with tokens: long documents, reasoning models and agents that loop are what push it up.
Data handling
OpenRouter's FAQ (opens in a new tab) says prompt/completion logging is off by default; request metadata is logged. Serving-provider policies are separate. data_collection: "deny" filters provider classifications; zdr: true filters zero-prompt-retention endpoints. Neither means all metadata vanishes. Training opt-out alone is not ZDR. Inspect the exact endpoint policy; these controls were not exercised here.
First request
Documentation-based suggestion — not run here.
Read first request
Sort a week of incoming emails. A three-person shop wants each email tagged invoice, order, complaint or other before anyone reads it.
Create an account, buy $10 of credits and create an API key with a $5 limit.
