How to choose a model through a unified AI gateway
A practical way to shortlist models, compare API access and validate a workload before switching an application to G0I.
Updated October 2, 2026 · G0I documentation team
Begin with a representative workload
Choose prompts that reflect what your application will actually do: short answers, long documents, code changes, extraction or tool calls. Keep a small evaluation set and define what a successful output looks like before comparing model names.
Compare capability and access separately
A published model is a candidate, not a guarantee that every endpoint, tool feature or subscription supports it. Inspect the catalog for model IDs and rates, then check the access plan and API format. Use pay-as-you-go where required instead of assuming a monthly plan includes every model.
Measure the cost of the whole task
Input and output tokens have separate prices. Long history, repeated context and retries can materially change the cost per completed task. Cache rates are relevant only when the endpoint supports caching and the request reports the corresponding usage.
Roll out incrementally
Start with a server-side key and a small request. Verify parsing, streaming completion and cancellation. Add bounded retries for transient failures, but treat invalid credentials and invalid inputs as configuration errors. Compare quality and your own measured latency before increasing traffic.
Continue with the documentation
Compare published models, review access plans, or start with the API quickstart.