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

# Available Models

DOS AI serves high-quality open-source LLMs via an OpenAI-compatible API. Self-hosted models run on dedicated RTX Pro 6000 GPUs with 96 GB VRAM in Asia-Southeast 1. Cloud models are served via partner providers for maximum coverage.

## Smart Routing

Use `dos-auto` as the model ID to let DOS AI automatically select the best model for each request. Smart routing uses a 15-dimension classifier to analyze your prompt and route to the optimal model based on task complexity, cost, and latency.

```python
response = client.chat.completions.create(
    model="dos-auto",  # Smart routing picks the best model
    messages=[{"role": "user", "content": "..."}],
)
```

## Model Catalog

### Self-Hosted (Lowest Latency)

| Model               | Provider | Context | Input      | Output     | Model ID |
| ------------------- | -------- | ------- | ---------- | ---------- | -------- |
| **Qwen3.5-35B-A3B** | Alibaba  | 128K    | $0.15 / 1M | $0.15 / 1M | `dos-ai` |

### Cloud Models

| Model                         | Provider         | Context | Input      | Output     | Model ID           |
| ----------------------------- | ---------------- | ------- | ---------- | ---------- | ------------------ |
| **Llama 4 Maverick 17B-128E** | Meta / DeepInfra | 1M      | $0.17 / 1M | $0.66 / 1M | `llama-4-maverick` |
| **Llama 4 Scout 17B-16E**     | Meta / DeepInfra | 640K    | $0.11 / 1M | $0.38 / 1M | `llama-4-scout`    |
| **DeepSeek V3**               | DeepSeek         | 128K    | $0.25 / 1M | $0.25 / 1M | `deepseek-v3`      |
| **Llama 3.3 70B**             | Meta             | 128K    | $0.20 / 1M | $0.20 / 1M | `llama-3.3-70b`    |
| **Llama 3.1 8B**              | Meta             | 128K    | $0.05 / 1M | $0.05 / 1M | `llama-3.1-8b`     |

> All prices are in USD. The catalog is DB-driven -- new models are added regularly. Check `GET /v1/catalog` or the [dashboard](https://app.dos.ai/models) for the latest list. See [Pricing](/models/pricing.md) for billing details.

### Embedding Models

| Model                  | Provider              | Dimensions | Model ID             |
| ---------------------- | --------------------- | ---------- | -------------------- |
| **Qwen3-Embedding-4B** | Alibaba / Self-hosted | 2560       | `qwen3-embedding-4b` |

See the [Embeddings API reference](/api-reference/embeddings.md) for the request/response format and code examples.

## Model Details

### Qwen3.5-35B-A3B (default)

Alibaba's Mixture-of-Experts model with 35 billion total parameters and 3 billion active parameters per forward pass. This architecture delivers excellent quality at remarkably low cost and latency, making it our **recommended default model** for most use cases.

* **Best for**: General-purpose chat, code generation, reasoning, multilingual tasks
* **Strengths**: Outstanding cost-efficiency, fast response times, strong multilingual support (especially CJK languages)
* **Model ID**: `dos-ai`

### Llama 4 Maverick 17B-128E

Meta's latest Mixture-of-Experts model with 17 billion active parameters and 128 experts. Strong reasoning and multilingual capabilities with an industry-leading 1 million token context window.

* **Best for**: Complex reasoning, long-context analysis, multilingual tasks
* **Strengths**: Massive context window, strong benchmark scores, efficient MoE architecture
* **Model ID**: `llama-4-maverick`

### Llama 4 Scout 17B-16E

Meta's efficient MoE model with 17 billion active parameters and 16 experts. Fast and cost-effective for everyday tasks with a 640K context window.

* **Best for**: Everyday tasks, fast responses, cost-sensitive workloads
* **Strengths**: Good balance of speed and quality, large context window
* **Model ID**: `llama-4-scout`

### DeepSeek V3

DeepSeek's latest Mixture-of-Experts model, known for strong performance across coding, math, and reasoning benchmarks.

* **Best for**: Code generation, mathematical reasoning, structured output
* **Strengths**: Competitive benchmark scores, good at structured/JSON output, strong code capabilities
* **Model ID**: `deepseek-v3`

### Llama 3.3 70B

Meta's 70-billion-parameter dense model. Offers top-tier reasoning and instruction-following capabilities.

* **Best for**: Complex reasoning, long-form content, detailed analysis
* **Strengths**: Strong English performance, excellent instruction following, robust safety tuning
* **Model ID**: `llama-3.3-70b`

### Llama 3.1 8B

Meta's efficient 8-billion-parameter model. An excellent choice when you need fast, affordable responses and the task does not require the full capability of a larger model.

* **Best for**: Simple tasks, high-throughput workloads, prototyping, cost-sensitive applications
* **Strengths**: Very low latency, lowest cost, suitable for classification and extraction tasks
* **Model ID**: `llama-3.1-8b`

## Choosing the Right Model

| Use Case                             | Recommended Model | Why                                            |
| ------------------------------------ | ----------------- | ---------------------------------------------- |
| Let DOS AI decide                    | `dos-auto`        | Smart routing picks the best model per request |
| General assistant / chatbot          | Qwen3.5-35B-A3B   | Best balance of quality, speed, and cost       |
| Long-context analysis (100K+ tokens) | Llama 4 Maverick  | 1M context window, strong reasoning            |
| Complex reasoning / analysis         | Llama 3.3 70B     | Dense model, top reasoning capability          |
| Code generation / math               | DeepSeek V3       | Top coding and math benchmark scores           |
| High-volume / low-cost tasks         | Llama 3.1 8B      | Fastest and cheapest option                    |
| Multilingual (CJK languages)         | Qwen3.5-35B-A3B   | Superior CJK language performance              |

## Listing Models via API

You can retrieve the current list of available models programmatically:

```bash
curl https://api.dos.ai/v1/models \
  -H "Authorization: Bearer YOUR_API_KEY"
```

For the full retail catalog with pricing and metadata:

```bash
curl https://api.dos.ai/v1/catalog \
  -H "Authorization: Bearer YOUR_API_KEY"
```

See the [Models API reference](/api-reference/models.md) for the full response schema.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://docs.dos.ai/models/available-models.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.