Pay per token. Nothing else.

Every model is priced at the vendor’s published list price or below, and there is no platform fee on top. Because we route across several providers for the same model, we can grant volume discounts per model. Each response tells you what the same tokens would have cost direct.

From the dashboard you pick a model you use a lot and ask for a better rate. We review the request, usually within a business day, and the approved discount applies to every call to that model from then on. Nothing to renegotiate with each vendor.

A provider only serves a model after our automated checks confirm it returns the right model, a correct answer and acceptable latency. Providers that fail or drift are removed from rotation automatically. You never call an unverified path.

Check it yourself. Open any model in the dashboard, click “Verify with your own key” and copy a two-line command. It sends the same prompts to the vendor’s own API with your vendor key and to OSMEXO with your OSMEXO key, then scores how alike the answers are: token counts, knowledge cutoff, a hidden-prompt probe and exact tasks. It runs on your machine and your keys go only to their own vendor. Before a route may serve anyone we run the same battery ourselves.

Yes. Point OpenCode, Cline, Roo Code, Continue or Aider at the OSMEXO base URL with your key and pick a model. Models marked “Coding ready” in the dashboard passed the workload those tools send: streamed tool calls, a tool round trip, files written through tools and a long file. The check runs per route and is repeated.

Only the base URL and the key. The gateway speaks the OpenAI wire format for chat, responses, embeddings, images, speech and transcription, so the official OpenAI SDKs and most frameworks work as they are.

Prepaid credits, from $5, paid in US dollars by PayPal or by Visa or Mastercard (debit or credit) through a hosted checkout; the card number never touches our servers. Other card brands are not accepted yet. There is no subscription and no seat fee, and credits never expire.

The gateway retries and falls back to the next verified provider for that model before the first token is sent, so your users see a slower answer instead of an error. You can see which route answered on every request.

Yes. Set a monthly limit for the organization, a daily or monthly budget per API key, and a low-balance alert. Purchases are capped per day and per month and can be raised on request. Unusual spend triggers an alert.

No. Prompts and completions pass through to the provider you are routed to and are stored only as much as you choose for your own request logs. Keys are stored hashed and secrets are never logged.