Call a hosted model
Set the base URL tohttps://api.runinfra.ai/v1, use a workspace API key, and pass a model id. There is nothing to provision.
deepseek-v4-flash, deepseek-v4-pro, qwen3-8-27b, qwen3-8-2-4t-a95b, and nemotron-3-5-lightning-30b. GET /v1/models is the authoritative list, and it reports when one is temporarily paused, which this page cannot.
Model APIs quickstart
Key, base URL, first call.
OpenAI compatibility
LangChain, LiteLLM, LlamaIndex, AI SDK, plain fetch.
Limits
Context, output budget, concurrency, tokens per minute.
Optimize your own model
Name a Hugging Face model in the chat, say what you want done with it, and the agent plans the work, prices it, measures candidates on real GPUs, and hands you the winner. The interface is the conversation, not a config file, so the request is a sentence:Run your first optimization
Fund, describe, run, keep the winner.
What accepting a plan locks
The accept boundary, and how to change your mind.
After the run
An endpoint you call, or a kit you run yourself.
Money, keys, and when it breaks
Pricing and credits
One balance, $10 minimum top-up, no subscription.
Account and access
Sign up, keys, workspaces, deletion.
Troubleshooting
401, 402, 403, 429, 503, and what to do about each.