One key for any model

OpenModel is a hosted model gateway that lets you call models from many providers with one API key. You top up one balance in USD, and Claude, GPT, and Gemini all run on that one key. You get one bill and one usage log. You never sign up with a provider or hold its keys.

An AI gateway is a single entry point in front of several model providers that also governs the traffic going through it. Yes, OpenModel is a hosted AI gateway. It works with the official Anthropic, OpenAI, and Google SDKs, retries a failed call for you, gives every API key its own USD quota, logs every request, and bills you once.

No code changes, and your SDK keeps working. Point your base URL at OpenModel, swap in an OpenModel API key, and your existing calls run as before. OpenModel takes four protocols: Anthropic Messages, OpenAI Responses, Gemini generateContent, and OpenAI Images. The official Anthropic, OpenAI, and Google SDKs work for the calls those protocols cover. If you are on Chat Completions today, move to Responses.

OpenModel checks your API key and its quota, then forwards the request. A streaming response starts coming back as soon as the model generates its first token, and any automatic retry finishes before that first token reaches you, so nothing cuts into a response in flight. The model catalog publishes each model's recent average latency and time to first token, so you can pick on real data.

No. OpenModel charges the catalog price in USD per million tokens, with no platform fee, per-request fee, subscription, or minimum charge, and a retried request is billed once. Four things move the rate: tiered pricing by input tokens, a time-based multiplier set by when a request finishes, cache reads and writes priced on their own, and client discounts. Multipliers can go above one, and the catalog shows them per model.

OpenModel retries for you. If a request errors or hits a rate limit, OpenModel automatically tries the same model again, and you pay only for the response you actually receive. Every retry finishes before the first token reaches you, so what you get is never stitched from two attempts. If every attempt fails, you get the provider's own error and nothing is charged by default.

Yes. OpenModel bills eligible models at the client discount when you call through Claude Code or Codex. Details are on the client benefits page.

Yes. Prompt caching keeps working when you call through OpenModel, and later requests in the same conversation can still hit a prefix you cached earlier. Cache reads and cache writes are priced on their own, at the cache prices in the catalog, and each shows up as its own line in your usage log. OpenModel works to keep your hit rate high.

No. OpenModel does not keep what you send or what comes back. Each request leaves one line in your usage log, and that line holds only metadata: model, token counts, cost, latency, status code, user agent, and IP address. Those lines are deleted after 180 days. OpenModel does not use what you send or receive to train models. Providers handle the requests they receive under their own policies.