Help
Have a question? Search here, or browse the tutorials, documents and FAQ below.
Tutorials
Step-by-step lessons
Documents
Guides, configuration and troubleshooting
FAQs
Frequently asked questions
What do 400, 401, 403, 429 and 500 errors mean?
400: bad request — check the model name and body. 401: the key is invalid or deleted. 403: the token has no access to that model or group, or your IP is not allowed. 429: too many requests — lower concurrency. 500/502/503: a temporary upstream problem — retry later; if it persists, send the time, model and request ID to the administrator.
What are group, model and completion ratios?
The model ratio sets the input price: input ($ per 1M tokens) = model ratio × 2. The completion ratio is output relative to input: output = input × completion ratio. The group ratio multiplies both. Turn on the Ratio switch on the Models page to see each model's ratios.
My balance is fine but requests say quota is exhausted
Tokens have their own quota cap. Check the token's remaining quota under API Tokens and raise it or set it to unlimited. A token quota is only a cap — it does not add to your balance.
Which base URL should I use?
Most OpenAI-style clients take https://17astra.com/v1. If the client adds /v1 itself (Cherry Studio, Claude Code), use https://17astra.com. Seeing /v1/v1 means one level too many.
Which models can search the web?
Models with a -search suffix look things up online before answering, for example gemini-2.5-pro-search. Search for “search” on the Models page to see them all.
Why can't a model state its exact version?
A model only knows its training data, which usually predates its own release, so it often names an older version. To see which model actually served a request, check the model name in Usage Logs.
Why are responses sometimes slow?
Upstream queues at peak hours, long contexts and reasoning time all add latency. Try a lighter model (flash, mini) or give the token several groups with cross-group retry on.