Models
JouleCloud serves open-weight models on its own GPU fleet. Below is the catalog available today. The full live list for your org is also visible in the console.
Served models
| Model | Params | Context | Good for | Status |
|---|---|---|---|---|
qwen2.5-7b-instruct | 7B | 32K (Qwen2.5 native) | General chat, drafting, extraction, RAG answers | available |
qwen2.5-3b-instruct | 3B | 32K (Qwen2.5 native) | High-throughput classification, cheap summarization | available |
Selecting a model
Set the model field in your request body to the model id, e.g. "model": "qwen2.5-7b-instruct". The router places the request on the cheapest healthy node serving that model. An unknown id returns a 400 error — see Errors.
Pricing
All served models currently share one usage-based rate: $0.20 per 1M input tokens and $0.60 per 1M output tokens. See Pricing for a worked example and the cost calculator.
Signed-out visitors see this curated served list. The live model catalog for your organization requires sign-in (it is queried with your org key).