> 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 and frontier cloud models via an OpenAI-compatible API. Self-hosted models run on dedicated GPU clusters in Singapore (`asia-southeast1-b`) with NVMe storage and sub-250ms TTFT. Cloud models are served via enterprise partner backends for maximum coverage and high availability.

## Smart Routing (`auto`)

Use `auto` (or `dos-auto`) as the model ID to let the DOS AI Smart Router automatically select the best model for each request. The router classifies incoming prompts into 3 complexity tiers (Simple, Medium, Complex) using task classification, reasoning depth, and context length.

```python
response = client.chat.completions.create(
    model="auto",  # Smart routing dynamically selects the optimal model
    messages=[{"role": "user", "content": "Analyze this multi-step agent workflow..."}],
)
```

### Escalation Hierarchy for Complex Requests

When a request requires advanced multi-step reasoning, complex coding, or autonomous agent loops, the Smart Router escalates to the lowest-priority active tier with verified tool calling and reasoning support:

| Priority        | Model                | Provider / Backend     | Context | Input (per 1M) | Output (per 1M) | Strengths                                            |
| --------------- | -------------------- | ---------------------- | ------- | -------------- | --------------- | ---------------------------------------------------- |
| **1 (Primary)** | **GPT-6 Luna**       | OpenAI / Azure Foundry | 1.05M   | **$0.10**      | **$0.50**       | Frontier efficiency, fast tool calling, agent loops  |
| **2**           | **Gemini 3.8 Flash** | Google Cloud           | 1M      | $0.79          | $3.94           | Sub-second TTFT, multimodal, massive context         |
| **3**           | **Qwen 3.8 Max**     | Alibaba Cloud          | 1M      | $2.00          | $6.00           | Complex bilingual logic, deep mathematical reasoning |

*Note: Simple and Medium tier requests remain on the in-house self-hosted model (`dos`) to optimize speed and cost.*

***

## Model Catalog

### Curated Platform Models & In-House Cluster (`dos`)

`dos` (or `dos-ai`) is DOS.AI's curated meta-model and intelligent execution endpoint, pre-tuned for DOSClaw agent workflows, ReAct loops, and native tool-calling with zero markup. Requests sent to `model: "dos"` default to the self-hosted Singapore GPU cluster (`qwen3.8-27b` at **$0.07** input / **$0.50** output per 1M tokens).

The platform also provides a **Curated Model Pool** of pre-selected, cost-efficient engines that can be invoked directly at each engine's exact wholesale list rate:

| Engine                  | Provider               | Context | Input Price (1M) | Output Price (1M) | Best For                                             |
| ----------------------- | ---------------------- | ------- | ---------------- | ----------------- | ---------------------------------------------------- |
| **Qwen 3.8 27B Dense**  | DOS.AI (Singapore GPU) | 256K    | **$0.07**        | **$0.50**         | In-house low-latency agent loops, sub-250ms TTFT     |
| **GPT-6 Luna**          | OpenAI / Azure         | 1.05M   | **$0.10**        | **$0.50**         | Complex reasoning, structured outputs, 1.05M context |
| **Gemini 3.8 Flash**    | Google Cloud           | 1M      | **$0.79**        | **$3.94**         | Ultra-fast multimodal processing, sub-second TTFT    |
| **DeepSeek V4.1 Flash** | DeepSeek / Alibaba     | 1M      | **$0.30**        | **$1.20**         | MoE high-throughput reasoning and coding logic       |

#### High-Availability & Circuit Breakers

Requests to `model: "dos"` execute on the in-house Singapore GPU cluster with automatic multi-provider circuit breaker failover defined in the Supabase catalog, ensuring 99.99% continuity during upstream provider maintenance.

***

### Frontier & Cloud Models

| Model                            | Provider           | Context | Input Price (1M) | Output Price (1M) | Model ID                     | Key Capabilities                                                        |
| -------------------------------- | ------------------ | ------- | ---------------- | ----------------- | ---------------------------- | ----------------------------------------------------------------------- |
| **GPT-6 Luna**                   | OpenAI / Azure     | 1.05M   | $0.10            | $0.50             | `gpt-6-luna`                 | Frontier reasoning, tools, 1.05M context                                |
| **GPT-6 Sol**                    | OpenAI / Azure     | 1.05M   | $2.00            | $10.00            | `gpt-6-sol`                  | Frontier flagship, deep architectural reasoning                         |
| **GPT-6 Astra**                  | OpenAI             | 1.05M   | $5.00            | $25.00            | `gpt-6-astra`                | Next-gen multistep engineering, async tools                             |
| **Gemini 3.8 Flash**             | Google Cloud       | 1M      | $0.79            | $3.94             | `gemini-3.8-flash`           | Hybrid reasoning, vision, sub-second TTFT                               |
| **DeepSeek V4.1 Flash**          | Alibaba / DeepSeek | 1M      | $0.30            | $1.20             | `deepseek-v4.1-flash`        | MoE, controllable thinking, 1M context                                  |
| **DeepSeek V4 Pro**              | DeepSeek / Alibaba | 1M      | $2.40            | $4.80             | `deepseek-v4-pro`            | Mathematical reasoning, complex agent workflows                         |
| **Qwen 3.8 Flash**               | Alibaba Cloud      | 1M      | $0.15            | $0.47             | `qwen3.8-flash`              | Ultra-fast lightweight reasoning, high-throughput chat                  |
| **Qwen 3.8 27B**                 | Alibaba Cloud      | 1M      | $0.50            | $3.00             | `qwen3.8-27b`                | Dense reasoning, tool calling, explicit caching                         |
| **Qwen 3.8 Max**                 | Alibaba Cloud      | 1M      | $2.00            | $6.00             | `qwen3.8-max`                | Multilingual reasoning, complex logic                                   |
| **Claude Sonnet 5**              | Anthropic          | 1M      | $3.00            | $15.00            | `claude-sonnet-5`            | Enterprise agentic coding, 128K output                                  |
| **Claude Opus 5**                | Anthropic          | 1M      | $15.00           | $75.00            | `claude-opus-5`              | Flagship reasoning, long-horizon tasks                                  |
| **Muse Wonder 1.0**              | Meta               | 128K    | $0.30            | $1.20             | `muse-wonder-1.0`            | Multimodal design, UI/UX & front-end code generation                    |
| **Muse Spark 1.3**               | Meta               | 128K    | $0.36            | $1.52             | `muse-spark-1.3`             | Flagship multimodal reasoning, native tool use, visual chain-of-thought |
| **Muse Spark 1.3 (Contributor)** | Meta               | 128K    | $0.00            | $0.00             | `muse-spark-1.3-contributor` | Complimentary contributor tier for multimodal reasoning & tools         |
| **Muse Spark 1.2**               | Meta               | 128K    | $0.26            | $1.04             | `muse-spark-1.2`             | Fast multimodal chat, vision & tool calling                             |
| **Muse Spark 1.1**               | Meta               | 128K    | $0.36            | $1.42             | `muse-spark-1.1`             | Foundational Muse Spark release with vision & structured outputs        |

> All prices are in USD and strictly adhere to DOS AI's **Zero-Markup Policy** (100% of provider published list price). Check `GET /v1/catalog` or the [web dashboard](https://app.dos.ai/models) for real-time status.

***

### Embedding Models

| Model                  | Provider              | Dimensions | Max Input Tokens | Model ID                 |
| ---------------------- | --------------------- | ---------- | ---------------- | ------------------------ |
| **Qwen3-Embedding-4B** | Alibaba / Self-hosted | 2560       | 32,768           | `qwen3-embedding-4b`     |
| **BGE-Large-EN-v1.5**  | Cloudflare / BAAI     | 1024       | 512              | `baai/bge-large-en-v1.5` |

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

***

### Image Generation Models

DOS.AI supports specialized text-to-image and visual generation models for autonomous agents, MCP marketing tools (`generate_marketing_image`), and direct visual generation:

| Model                            | Provider      | Resolution / Tiers | Billing Unit                     | Model ID           | Primary Use Case                                           |
| -------------------------------- | ------------- | ------------------ | -------------------------------- | ------------------ | ---------------------------------------------------------- |
| **Wan 2.2 T2I Flash**            | Alibaba Cloud | Up to 1024×1024    | Per Image ($0.08)                | `wan2.2-t2i-flash` | High-speed creative marketing visual generation, banners   |
| **Gemini Image (Nano Banana 2)** | Google Cloud  | 1K / 2K            | Variable Tokens (Input + Output) | `nano-banana-2`    | Multimodal instruction-following, bilingual marketing copy |
| **Nano Banana Pro**              | Google Cloud  | 1K / 2K            | Variable Tokens (Input + Output) | `nano-banana-pro`  | High-fidelity photorealistic rendering and design assets   |

> **Agent Model Priority Rule**: In DOSClaw, image generation models can be included in an agent's ordered **Model Priority list** (`model_chain`) to designate preferred visual tool routes. However, an image model **cannot** be assigned as the Primary Chat Model (`model_chain[0]`), which strictly requires conversational chat capabilities.

***

## Model Details

### GPT-6 Luna

OpenAI's latest cost-efficient frontier reasoning model. Features state-of-the-art instruction following, coding, structured JSON outputs, and native tool-calling agent capabilities at more than 50% lower cost than previous generations ($0.10 in / $0.50 out).

* **Best for**: Autonomous agent loops, function calling, high-volume production chat, code assistance
* **Strengths**: 1.05M context window, exceptional price-to-performance, Azure Foundry low-latency hosting
* **Model ID**: `gpt-6-luna`

### DOS Curated Meta-Model (`dos`)

DOS.AI's curated meta-model pre-tuned for DOSClaw agent workflows, ReAct loops, and low-latency execution. Served from our in-house Singapore GPU cluster (sub-250ms TTFT) with unified zero-markup billing.

* **Best for**: Agent turn executions, customer support, fast interactive chat, CJK bilingual tasks
* **Strengths**: Sub-250ms TTFT, 256K context, pre-selected 4-engine curated pool, automatic catalog circuit breakers
* **Model ID**: `dos`, `dos-ai`

### Gemini 3.8 Flash

Google's high-speed multimodal reasoning model with hybrid thinking, 1M context window, and sub-second execution.

* **Best for**: Real-time document parsing, large multimodal payloads, high-throughput applications
* **Strengths**: 1M context, sub-second TTFT, native audio/video/image comprehension
* **Model ID**: `gemini-3.8-flash`

### Meta Muse Family (`muse-wonder-1.0` & `muse-spark-1.3`)

Meta Superintelligence Labs' flagship natively multimodal family built for visual chain-of-thought reasoning, multi-agent orchestration, and high-craft front-end/UI engineering.

* **Best for**: UI/UX design & interactive front-end code generation (`muse-wonder-1.0`), multimodal reasoning & tool calling (`muse-spark-1.3`)
* **Strengths**: 128K context window, native vision + text input, 89%–90% prompt caching discount ($0.04/1M cached input), complimentary `muse-spark-1.3-contributor` tier
* **Model ID**: `muse-wonder-1.0`, `muse-spark-1.3`, `muse-spark-1.3-contributor`, `muse-spark-1.2`, `muse-spark-1.1`

***

## Choosing the Right Model

| Use Case                                     | Recommended Model  | Why                                                                      |
| -------------------------------------------- | ------------------ | ------------------------------------------------------------------------ |
| **Let DOS AI decide**                        | `auto`             | Smart Router dynamically matches task complexity and cost                |
| **Fastest agent execution & lowest latency** | `dos`              | Self-hosted 27B dense cluster in Singapore, sub-250ms TTFT               |
| **Ultra-low cost frontier reasoning**        | `gpt-6-luna`       | $0.10 / $0.50 per million tokens with 1.05M context                      |
| **Sub-second large document analysis**       | `gemini-3.8-flash` | 1M context window, Google Cloud high-throughput backend                  |
| **UI/UX design & front-end code generation** | `muse-wonder-1.0`  | Meta's specialized design & creative coding model ($0.30 / $1.20 per 1M) |
| **Enterprise agentic coding & architecture** | `claude-sonnet-5`  | Frontier coding quality with 128K max output tokens                      |
| **Deep mathematical reasoning**              | `deepseek-v4-pro`  | SOTA chain-of-thought problem solving                                    |

***

## Operational Telemetry & Provider Comparison

Every model detail page in the [Dashboard](https://app.dos.ai/models) features live, rolling operational telemetry across all active provider routes:

| Metric                         | Description                                  | Source / Contract                                             |
| ------------------------------ | -------------------------------------------- | ------------------------------------------------------------- |
| **Median Latency (p50)**       | Median round-trip execution latency          | Real observed inference requests over a 7-day rolling window  |
| **P95 Latency**                | 95th-percentile response latency             | Measures tail latency under peak conditions                   |
| **TTFT (Time-to-First-Token)** | Time to first streaming token                | Observed on streaming requests across edge locations          |
| **Throughput (TPS)**           | Tokens generated per second (`tok/s`)        | Real measured generation throughput                           |
| **Uptime & Reliability**       | Success ratio percentage over total attempts | Excludes client-side cancellations and user-side bad requests |

> **Telemetry Sample Threshold**: To maintain rigorous evidence standards, a model route requires at least **20 valid samples** before publishing live operational telemetry. Routes with fewer observations report `insufficient_data` or `unavailable` with clean dash (`-`) indicators until warmed up.

***

## Independent Benchmarks

DOS AI continuously synchronizes independent benchmark evidence from leading evaluation authorities, attributing each metric with capture dates and direct evaluation links:

* **Artificial Analysis**: Intelligence Index, Coding Index, and Agentic Index.
* **Design Arena**: Elo rating, win rate percentage, and leaderboard rank.

Benchmark scores are displayed directly on each model's detail page, complete with normalized visual progress bars (0–100) and direct hyperlinks to public methodology audits.

***

## Listing Models via API

Retrieve the current list of available models programmatically:

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

For the full retail catalog including pricing, context lengths, and capability flags:

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


---

# 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 by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

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

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

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.