# DeepSeek OCR — Atlas Cloud API

> The latest Deepseek model.

This is the machine-readable API reference for **DeepSeek OCR** on Atlas Cloud,
a unified API platform for 400+ AI models across text, image, video, audio and 3D.

- **Model ID**: `deepseek-ai/deepseek-ocr`
- **Built by**: DeepSeek
- **Modality**: Text
- **Context length**: 8,192 tokens
- **Max output tokens**: 8,192
- **Model page**: https://www.atlascloud.ai/models/deepseek-ai/deepseek-ocr
- **API key**: https://www.atlascloud.ai/console/api-keys
- **Docs**: https://www.atlascloud.ai/docs

## Pricing on Atlas Cloud

- **Input**: $0.04 per 1M tokens
- **Output**: $0.08 per 1M tokens
- Pay-as-you-go. No minimum spend, no subscription required.

> **These are the authoritative Atlas Cloud rates for this model.** Any price that
> appears in the vendor description further down refers to a different platform or
> a different model variant and does not apply here.

## Use this model from an AI agent

Atlas Cloud ships three first-party integration surfaces. All three authenticate
with the same API key via the `ATLASCLOUD_API_KEY` environment variable.

### MCP server

The official MCP server (`atlascloud-mcp`) exposes this model to any
MCP-compatible host — Claude Code, OpenAI Codex, Cursor, Gemini CLI, Goose,
Claude Desktop. One-line install:

```bash
# Claude Code
claude mcp add atlascloud -- npx -y atlascloud-mcp

# OpenAI Codex CLI
codex mcp add atlascloud -- npx -y atlascloud-mcp

# Gemini CLI
gemini mcp add atlascloud -- npx -y atlascloud-mcp

export ATLASCLOUD_API_KEY="your-api-key"
```

Then ask in plain English; the agent calls `atlas_chat` with `model: "deepseek-ai/deepseek-ocr"`.
The server fetches each model's schema and validates parameters before submitting,
so invalid requests fail fast without spending credits.

MCP docs: https://www.atlascloud.ai/docs/mcp-server

### Agent Skills

`atlas-cloud-skills` is a portable skill package (API reference, code templates in
Python / Node.js / cURL, model IDs with pricing) for Claude Code, Cursor, Codex and
12+ other agents:

```bash
npx skills add AtlasCloudAI/atlas-cloud-skills
export ATLASCLOUD_API_KEY="your-api-key"
```

Skills docs: https://www.atlascloud.ai/docs/skills

### CLI

The `atlas` binary runs Atlas Cloud from a terminal or CI script. Async media jobs
are polled and downloaded automatically (use `--no-download` when a script only
needs the output URLs):

```bash
# Install (Homebrew, npm, or shell installer)
brew install AtlasCloudAI/tap/atlascloud
# npm install -g atlascloud-cli
# curl -fsSL https://raw.githubusercontent.com/AtlasCloudAI/cli/main/install.sh | sh

atlas auth login
atlas chat "Your prompt here" --model deepseek-ai/deepseek-ocr
```

CLI docs: https://www.atlascloud.ai/docs/cli

## HTTP API reference

This model is served over an **OpenAI-compatible** endpoint. Any OpenAI SDK works —
just override the base URL and use the Atlas Cloud model ID.

```http
POST https://api.atlascloud.ai/v1/chat/completions
Authorization: Bearer $ATLASCLOUD_API_KEY
Content-Type: application/json
```

### curl

```bash
curl -X POST "https://api.atlascloud.ai/v1/chat/completions" \
  -H "Authorization: Bearer $ATLASCLOUD_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "deepseek-ai/deepseek-ocr",
  "messages": [
    {
      "role": "user",
      "content": [
        {
          "type": "image_url",
          "image_url": {
            "url": "data:image/png;base64,<BASE64_IMAGE_DATA>"
          }
        },
        {
          "type": "text",
          "text": "Please describe the content of this image"
        }
      ]
    }
  ],
  "max_tokens": 4096,
  "temperature": 0.7,
  "stream": false
}'
```

### Python (OpenAI SDK)

```python
import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["ATLASCLOUD_API_KEY"],
    base_url="https://api.atlascloud.ai/v1",
)

response = client.chat.completions.create(
    model="deepseek-ai/deepseek-ocr",
    messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)
```

### Multimodal input

This model accepts image input alongside text. Media parts go inside `messages[].content[]`.

- **Image** — part `type: "image_url"`, with `image_url.url` set to a public URL or a `data:image/...;base64,` URI.
  - Formats: image/png, image/jpeg, image/webp, image/gif

**Example request body**:

```json
{
  "model": "deepseek-ai/deepseek-ocr",
  "messages": [
    {
      "role": "user",
      "content": [
        {
          "type": "image_url",
          "image_url": {
            "url": "data:image/png;base64,<BASE64_IMAGE_DATA>"
          }
        },
        {
          "type": "text",
          "text": "Please describe the content of this image"
        }
      ]
    }
  ],
  "max_tokens": 4096,
  "temperature": 0.7,
  "stream": false
}
```

### Supported sampling parameters

`temperature`, `top_p`, `top_k`, `min_p`, `frequency_penalty`, `presence_penalty`, `repetition_penalty`, `stop`, `seed`, `max_tokens`, `logit_bias`

Context window: 8,192 tokens.

## About this model

_Vendor-supplied description. Any pricing or endpoint mentioned below refers to_
_other platforms — use the Atlas Cloud values above._

### DeepSeek-V3.1

#### Introduction

DeepSeek-V3.1 is a hybrid model that supports both thinking mode and non-thinking mode. Compared to the previous version, this upgrade brings improvements in multiple aspects:

- **Hybrid thinking mode**: One model supports both thinking mode and non-thinking mode by changing the chat template. 

- **Smarter tool calling**: Through post-training optimization, the model's performance in tool usage and agent tasks has significantly improved.

- **Higher thinking efficiency**: DeepSeek-V3.1-Think achieves comparable answer quality to DeepSeek-R1-0528, while responding more quickly.

DeepSeek-V3.1 is post-trained on the top of DeepSeek-V3.1-Base, which is built upon the original V3 base checkpoint through a two-phase long context extension approach, following the methodology outlined in the original DeepSeek-V3 report. We have expanded our dataset by collecting additional long documents and substantially extending both training phases. The 32K extension phase has been increased 10-fold to 630B tokens, while the 128K extension phase has been extended by 3.3x to 209B tokens. Additionally, DeepSeek-V3.1 is trained using the UE8M0 FP8 scale data format to ensure compatibility with microscaling data formats.

#### Model Downloads

| **Model** | **#Total Params** | **#Activated Params** | **Context Length** | **Download** |
| :------------: | :------------: | :------------: | :------------: | :------------: |
| DeepSeek-V3.1-Base | 671B | 37B | 128K | [HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-V3.1-Base) \| [ModelScope](https://modelscope.cn/models/deepseek-ai/DeepSeek-V3.1-Base) |
| DeepSeek-V3.1 | 671B | 37B | 128K | [HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-V3.1) \| [ModelScope](https://modelscope.cn/models/deepseek-ai/DeepSeek-V3.1) |

#### Chat Template

The details of our chat template is described in `tokenizer_config.json` and `assets/chat_template.jinja`. Here is a brief description.

_(Description truncated. Full text on the model page.)_

---

Atlas Cloud — one API for 400+ AI models. Model page: https://www.atlascloud.ai/models/deepseek-ai/deepseek-ocr · Docs: https://www.atlascloud.ai/docs