> For the complete documentation index, see [llms.txt](https://docs.trezalabs.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.trezalabs.com/pipelines/models.md).

# Models

Every generation node in a pipeline runs a model. Treza reaches models through an OpenAI-compatible backend that spans providers like Together, OpenRouter, Fireworks, and DeepSeek, so you can pick the best model for each step and never get locked into one vendor.

***

## The catalog

The catalog covers every modality. Pick a model per node from the lists below, or type any model id straight in (see [Any model id](#any-model-id)).

### Video

Cinematic clips from a sentence, with duration and aspect ratio set per node.

* Veo 3.1
* Veo 3.1 Fast
* Sora 2 Pro

### Image

Product shots, illustrations, and campaign art with full response metadata.

* Gemini 2.5 Flash Image
* Gemini 3 Pro Image

### Language

Drafting, rewriting, classification, and reasoning with open-weight models.

* Llama 3.3 70B
* DeepSeek R1
* Qwen2.5 72B
* Mixtral 8x22B

The catalog also covers the major open families: Llama, Mistral, Qwen, DeepSeek, Gemma, and Phi.

### Audio & embeddings

Transcribe speech and vectorize text for search and retrieval steps.

* Whisper Large v3
* BGE Large EN

***

## Any model id

Not listed? Type any Hugging Face or OpenRouter model id directly into the node and Treza will route to it through the OpenAI-compatible backend. This means new models are usable the day they ship, without waiting for them to be added to the catalog.

***

## Swap models, no lock-in

* **One model per node**, changeable at any time. Swapping a model never requires rewriting the rest of the pipeline.
* **Per-node choice.** Use a fast, cheap model to refine a brief and a premium model to render the final asset.
* **Bring your own keys.** Store provider API keys in the encrypted [secrets manager](/getting-started/concepts.md#secrets). Keys are used server-side and never returned by any API, so you are never tied to one price or one vendor.

***

## Cost

Model runs are paid for with prepaid credits, and only successful runs are charged. There is no subscription: you buy credits once and spend them as you run models. Every run records its token counts and cost in [run history](/pipelines/publishing.md#run-history). See [Plans & Pricing](/ai-gateway/plans-and-pricing.md).