Model catalog
The platform keeps four independent lists: chat models, embedding models, rerankers and media models (speech and images). No keys are needed — all of it works right after you register, and every model’s price is declared up front, next to its name.
The lists move with the market: new models turn up in the dropdowns without you doing anything. How many chat models are available right now is visible under Settings → Models.
Where a model is picked
| What you configure | Where |
|---|---|
| Chat model | the Model field of the llm node — a config node wired with a link_llm edge into llm_response, ai_agent or structured_output_llm |
| Embeddings | the embedding node and the model choice when a knowledge base is created |
| Reranker | the rerank node |
| Speech and images | the image_generate, text_to_speech, speech_to_text nodes |
| Web search | the websearch node |
On the media nodes the set of fields depends on the model picked: the form rebuilds itself around it, because speech has a voice and an image has a size and a quality.
A fresh node opens on the default model — inexpensive, and strong enough for a first bot to answer sensibly. There is no need to change it to get started.
Visibility and price
Settings → Models decides which models appear in the dropdowns at all. They are grouped by source — platform models, workspace models and fallback chains — and each shows its price in quarks per 1000 tokens, input and output separately.
Models
Platform models
A disabled model disappears from the dropdowns
| Model | Quarks per 1000 tokens | Show |
|---|---|---|
| GPT-6 Luna | 0.1 / 0.5 | on |
| Qwen3 8B | 0.117 / 0.455 | on |
| Nemotron 3 Ultra (free) | 0 / 0 | off |
Turning models off is worth doing once a team works on the scenarios: a list of hundreds becomes a short one with nothing surplus and nothing too expensive in it.
Embeddings and knowledge bases
A knowledge base’s vector size is fixed once, at creation. From then on it beats any change in the model list — otherwise a working base would turn into a pile of incompatible vectors. So the embedding model of an existing base cannot be changed: create a new base and load the documents into it.