> ## Documentation Index
> Fetch the complete documentation index at: https://docs.salad.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Qwen3.5-35B-A3B with llama.cpp Recipe

> Serve Qwen3.5-35B-A3B with llama.cpp as an OpenAI-compatible API for OpenClaw, OpenCode, and other tools.

*Last Updated: March 23, 2026*

<Tip>Deploy from the [SaladCloud Portal](https://portal.salad.com).</Tip>

## Overview

This recipe runs `Qwen3.5-35B-A3B` with the official [llama.cpp](https://github.com/ggml-org/llama.cpp) CUDA server on a
Salad GPU. The model downloads automatically on first startup, the built-in llama.cpp web UI is available at your
deployment URL, and the container exposes an OpenAI-compatible API for tools such as OpenClaw, OpenCode, and other
compatible clients.

This recipe is designed to be easy for nontechnical users:

* the model is already chosen for you
* it is public by default, so you can test it immediately after deployment
* thinking is enabled by default
* you can start with the built-in web UI, then connect other tools later

## Quick Start

1. Open the [SaladCloud Portal](https://portal.salad.com).
2. Deploy the **Qwen3.5-35B-A3B (llama.cpp)** recipe.
3. Enter a **Container Group Name**.
4. Decide whether to enable **Require Container Gateway Authentication**:
   * Disabled: public access.
   * Enabled: requests must include your SaladCloud API key.
5. Choose whether to keep **Enable Thinking / Reasoning** turned on.
6. Deploy and wait for the first startup to finish.

<Callout variation="note">
  Model is downloaded at startup, so it can take up to about 20 minutes before the deployment becomes ready.
</Callout>

Once the container is ready, you can either open the built-in UI in a browser or connect an OpenAI-compatible client to
`/v1/chat/completions`.

## Use With OpenClaw

If you want to connect this recipe to OpenClaw, follow this guide:

* [Use OpenClaw with a Salad-hosted LLM](/container-engine/how-to-guides/openclaw/openclaw-ollama-salad-hosted-telegram)

## Defaults

The recipe comes preconfigured with these defaults:

* Model source: `unsloth/Qwen3.5-35B-A3B-GGUF:UD-Q4_K_XL`
* Model alias: `qwen3.5-35b-a3b`
* Context size: `262144`
* Parallel slots: `1`
* Thinking: enabled by default
* Sampling defaults: `temperature 0.6`, `top_p 0.95`, `min_p 0.0`, `top_k 20`
* Authentication: disabled by default

`temperature`, `top_p`, and `min_p` are startup defaults. You can still override them per request in your inference
payload.

## Thinking Mode

When thinking is enabled, you can control it per request:

* Add `/think` to explicitly enable reasoning for that turn.
* Add `/no_think` to disable reasoning for that turn.

If you disable thinking in the deployment form, the recipe sets Qwen's hard switch and `/think` will no longer override
it.

## Authentication

**Require Container Gateway Authentication** is available in the deployment form and is unchecked by default.

* Disabled: anyone with the URL can call the API.
* Enabled: every request must include the `Salad-Api-Key` header.

If you enable authentication, see [Sending Requests](/container-engine/how-to-guides/gateway/sending-requests) for the
header format.

## Example Request

```bash theme={null}
curl https://<your-dns>.salad.cloud/v1/chat/completions \
  -X POST \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "qwen3.5-35b-a3b",
    "messages": [
      {"role": "system", "content": "You are a helpful assistant."},
      {"role": "user", "content": "Write a short explanation of mixture-of-experts models."}
    ],
    "max_tokens": 256
  }'
```

If you enabled authentication during deployment, add:

```bash theme={null}
-H 'Salad-Api-Key: <api-key>'
```

## For Technical Users

If you want to tune llama.cpp later, open the container group in the SaladCloud Portal and edit **Advanced
Configuration**.

Useful environment variables include:

* `LLAMA_ARG_HF_REPO` to change the Hugging Face GGUF repo
* `LLAMA_ARG_MODEL_URL` to point directly to a `.gguf` file
* `LLAMA_ARG_CTX_SIZE` to change the context window
* `LLAMA_ARG_N_GPU_LAYERS` to control GPU offload
* `LLAMA_ARG_N_PARALLEL` to change concurrency

For full llama.cpp server options, see:

* [llama.cpp server docs](https://github.com/ggml-org/llama.cpp/tree/master/tools/server)
* [Server args and environment variable mapping](https://github.com/ggml-org/llama.cpp/blob/master/tools/server/README.md#usage)

## Source Code

* [<Icon icon="github" size="24" /> Recipe Source](https://github.com/SaladTechnologies/salad-recipes/tree/master/recipes/qwen3.5-35b-a3b-llama-cpp)
* [Unsloth Qwen3.5 model notes](https://unsloth.ai/docs/models/qwen3.5#qwen3.5-35b-a3b)
* [llama.cpp Project](https://github.com/ggml-org/llama.cpp)