Models & model IDs
Where to find model IDs and prices, and how to pass them in requests.
Every request names a model in its model field. Use the exact ID shown in the catalog.
Finding models
- Browse: the model catalog lists every model with its provider, input and output price per million tokens, and the endpoints it supports.
- Programmatically: call List models with your key.
curl https://api.cheapaimodel.ai/v1/models \
-H "Authorization: Bearer $CHEAPAIMODEL_API_KEY"Which endpoint for which model
Chat models are called through /v1/chat/completions, /v1/responses or /v1/messages, depending on what the model supports. Embedding models are only available through /v1/embeddings. The catalog shows the supported endpoints for each model; a model is supported only on the endpoints it lists.
Examples in these docs use deepseek-v4-1-flash and glm-5.3 for illustration. Always check the catalog for the current list.
Sampling limits
Some models accept only fixed values for their sampling parameters and reject any other value with 400:
| Model | temperature | top_p |
|---|---|---|
kimi-k3 | 1 or 0.6 | 0.95 |
The safest choice is to leave both parameters out, so the model uses its own defaults. If your tool always sends them, set exactly these values. Moonshot's model overview fixes kimi-k3 at temperature 1.0 and top_p 0.95; 0.6 also passes through our gateway.
kimi-k3 also accepts only n = 1, and presence_penalty and frequency_penalty only at 0.
Tools that send them by default
These tools send a temperature or top_p even when you don't set one, so their defaults fail on kimi-k3:
| Tool | What it sends | What to set |
|---|---|---|
| Fabric | temperature 0.7, top_p 0.9 | Add --raw to send no sampling options, or pass -t 1 -T 0.95. |
| RAGFlow | The temperature and top_p of its parameter presets and agent defaults | In the model's parameters, set Temperature to 1 and Top P to 0.95, or switch both off. |
| LlamaIndex | temperature 0.1 | Pass temperature=1.0 to OpenAILike. |
For any other tool, look for its temperature and top_p settings and either clear them or set the values above.