Call open models like you call OpenAI
Prepaid balance, per-token billing, no infrastructure to run. Change three settings in your existing OpenAI code and every request becomes traceable and explainable.
Design target for the time the gateway adds on top of the upstream first token. Measured latency per model is on each model page.
Monthly, backed by service credits.
Published prices rise only after 30 days.
By email and response header.
Built for running in production
base_url="https://gridport.ai/v1",api_key=os.environ["GRIDPORT_API_KEY"],model="deepseek-ai/DeepSeek-V4-Flash",Parameters the platform does not know are ignored and named in the x-tf-ignored-params response header instead of failing the call. A parameter that needs a capability the model lacks, such as tools on a model without tool calling, is refused with 400 unsupported_parameter.
- 1route-aTimed out at 800 ms · not billed
- 2route-bServed · switched before the first byte
Where a model has more than one route, a failure before the first byte switches route transparently and writes the switch to the response headers. Each model page shows how many routes it has.
This was not a problem with your request.
Every response carries a Request ID. The log shows the error reason, the routing trail across attempts, tokens and cost for 90 days.
Model and IP allowlists, per-key budgets, project budgets and three-stage balance alerts. Every block returns a specific error code.
Repeated prefixes are billed at the model's cached input rate, listed on each model page. Cache state is reported per request.
Try any model with real parameters, see first-token latency and cost per reply, then export the exact request as code.
What happens to a request
Every call goes through the same pipeline. Each step either passes the request on or returns a specific error code you can act on.
- 01/Authenticate
The key is hashed and checked against project, organization and model allowlists. A disabled key stops working within five seconds.
- 02/Limit and check balance
Rate limits apply per key, project, organization and tier — the strictest wins. Available balance is checked before any upstream call.
- 03/Route
The model resolves to a version, then to a route. If the first route fails before the first byte, the gateway switches and records the attempt.
- 04/Meter
Input, cached input and output are metered separately at the price in effect when the request started, then settled into the ledger.
Per million tokens
Prepaid balance. Cached input is billed at a lower rate. Batch jobs use the batch price: half the standard rate unless the model lists its own.
Common questions
Three settings change: base_url, api_key and model. Nothing else in your code changes. The migration assistant checks a pasted request parameter by parameter and suggests a model mapping.
Your first call in ten minutes
Create a project key and compare actual usage before committing to a plan.