DeepSeek V4 Pro

Model ID

Use with the Workers AI binding, the AI Gateway REST API, or the Unified API.

Calling the provider directly? Use the provider's model name instead, including with provider-native endpoints or the provider's own SDK. For example, Anthropic's name for anthropic/claude-opus-4.8 is claude-opus-4-8.

Third-party

DeepSeek V4 Pro is a high-capability reasoning model from DeepSeek, served via Fireworks infrastructure for production-grade inference.

Model Info
Context Window ↗	131,072 tokens
More information	link ↗
Request formats	Chat Completions
Pricing	Input (per 1M tokens)$1.74 Output (per 1M tokens)$3.48 Cached input (per 1M tokens)$0.145

Usage

const response = await env.AI.run(
  'deepseek/deepseek-v4-pro',
  {
    messages: [{ content: 'What is the capital of France?', role: 'user' }],
    model: 'deepseek/deepseek-v4-pro',
  },
)
console.log(response)

curl https://api.cloudflare.com/client/v4/accounts/$CLOUDFLARE_ACCOUNT_ID/ai/v1/chat/completions \
  --header "Authorization: Bearer $CLOUDFLARE_API_TOKEN" \
  --header "Content-Type: application/json" \
  --data '{
  "model": "deepseek/deepseek-v4-pro",
  "messages": [
    {
      "content": "What is the capital of France?",
      "role": "user"
    }
  ]
}'

The capital of France is **Paris**.

{
  "id": "chatcmpl-3a08845344c942108c3ab7b29112f012",
  "object": "chat.completion",
  "created": 1781047641,
  "model": "deepseek/deepseek-v4-pro",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "The capital of France is **Paris**.",
        "reasoning_content": "We need to answer the question: \"What is the capital of France?\" This is straightforward. The capital of France is Paris. I should answer concisely."
      },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 11,
    "completion_tokens": 43,
    "total_tokens": 54,
    "prompt_tokens_details": {
      "cached_tokens": 0
    }
  },
  "gatewayMetadata": {
    "keySource": "Unified"
  }
}

Examples

With System Message — Using a system message to set context

const response = await env.AI.run(
  'deepseek/deepseek-v4-pro',
  {
    messages: [
      { content: 'You are a helpful coding assistant specializing in Python.', role: 'system' },
      { content: 'How do I read a JSON file in Python?', role: 'user' },
    ],
    model: 'deepseek/deepseek-v4-pro',
    temperature: 0.3,
  },
)
console.log(response)

curl https://api.cloudflare.com/client/v4/accounts/$CLOUDFLARE_ACCOUNT_ID/ai/v1/chat/completions \
  --header "Authorization: Bearer $CLOUDFLARE_API_TOKEN" \
  --header "Content-Type: application/json" \
  --data '{
  "model": "deepseek/deepseek-v4-pro",
  "messages": [
    {
      "content": "You are a helpful coding assistant specializing in Python.",
      "role": "system"
    },
    {
      "content": "How do I read a JSON file in Python?",
      "role": "user"
    }
  ],
  "temperature": 0.3
}'

To read a JSON file in Python, you use the built-in `json` module. The most common approach is `json.load()` which reads directly from a file object.

### Basic example
```python
import json

with open('data.json', 'r', encoding='utf-8') as f:
    data = json.load(f)

print(data)
```

### Key points
- **`json.load(file)`** – parses JSON from a file-like object.
- **`json.loads(string)`** – parses JSON from a string (useful if you already have JSON data in memory).
- Always open the file in read mode (`'r'`) and specify the correct encoding (usually `'utf-8'`).
- The result is a Python dictionary (if the JSON is an object) or a list (if it’s an array).

### Handling errors
Wrap the loading in a `try`/`except` to catch malformed JSON or file issues:
```python
import json

try:
    with open('data.json', 'r', encoding='utf-8') as f:
        data = json.load(f)
except FileNotFoundError:
    print("File not found.")
except json.JSONDecodeError as e:
    print(f"Invalid JSON: {e}")
```

### Reading from a string
If you already have a JSON string:
```python
json_string = '{"name": "Alice", "age": 30}'
data = json.loads(json_string)
```

That’s it! Let me know if you need help with writing JSON or more advanced usage.

{
  "id": "chatcmpl-1aecff3044dc4e349362b9237f0e63b4",
  "object": "chat.completion",
  "created": 1781047642,
  "model": "deepseek/deepseek-v4-pro",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "To read a JSON file in Python, you use the built-in `json` module. The most common approach is `json.load()` which reads directly from a file object.\n\n### Basic example\n```python\nimport json\n\nwith open('data.json', 'r', encoding='utf-8') as f:\n    data = json.load(f)\n\nprint(data)\n```\n\n### Key points\n- **`json.load(file)`** – parses JSON from a file-like object.\n- **`json.loads(string)`** – parses JSON from a string (useful if you already have JSON data in memory).\n- Always open the file in read mode (`'r'`) and specify the correct encoding (usually `'utf-8'`).\n- The result is a Python dictionary (if the JSON is an object) or a list (if it’s an array).\n\n### Handling errors\nWrap the loading in a `try`/`except` to catch malformed JSON or file issues:\n```python\nimport json\n\ntry:\n    with open('data.json', 'r', encoding='utf-8') as f:\n        data = json.load(f)\nexcept FileNotFoundError:\n    print(\"File not found.\")\nexcept json.JSONDecodeError as e:\n    print(f\"Invalid JSON: {e}\")\n```\n\n### Reading from a string\nIf you already have a JSON string:\n```python\njson_string = '{\"name\": \"Alice\", \"age\": 30}'\ndata = json.loads(json_string)\n```\n\nThat’s it! Let me know if you need help with writing JSON or more advanced usage.",
        "reasoning_content": "We need to provide a clear, concise answer on how to read a JSON file in Python. The user likely wants to know the standard method using the `json` module. We'll explain opening the file, using `json.load()` for file objects, and `json.loads()` for strings. Also mention error handling, encoding, and maybe a simple example. Keep it friendly and informative."
      },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 24,
    "completion_tokens": 413,
    "total_tokens": 437,
    "prompt_tokens_details": {
      "cached_tokens": 0
    }
  },
  "gatewayMetadata": {
    "keySource": "Unified"
  }
}

Streaming Response — Enable streaming for real-time output

const response = await env.AI.run(
  'deepseek/deepseek-v4-pro',
  {
    messages: [{ content: 'Explain the concept of recursion with a simple example.', role: 'user' }],
    model: 'deepseek/deepseek-v4-pro',
    stream: true,
  },
)
console.log(response)

curl https://api.cloudflare.com/client/v4/accounts/$CLOUDFLARE_ACCOUNT_ID/ai/v1/chat/completions \
  --header "Authorization: Bearer $CLOUDFLARE_API_TOKEN" \
  --header "Content-Type: application/json" \
  --data '{
  "model": "deepseek/deepseek-v4-pro",
  "messages": [
    {
      "content": "Explain the concept of recursion with a simple example.",
      "role": "user"
    }
  ],
  "stream": true
}'

Recursion is a programming technique where a function calls itself to solve a smaller version of the same problem. Each recursive call works on a simpler input, and there’s always a **base case** that stops the recursion, preventing an infinite loop.

### The Two Essential Parts
1. **Base case** – the simplest scenario that can be answered directly (no more recursive calls).
2. **Recursive case** – the function calls itself with a smaller/simpler argument, moving toward the base case.

### Simple Example: Factorial
The factorial of a non-negative integer `n` (written `n!`) is the product of all positive integers up to `n`.  
Definition:  
- `0! = 1` (base case)  
- `n! = n × (n−1)!` for `n > 0` (recursive case)

#### Python implementation
```python
def factorial(n):
    if n == 0:          # base case
        return 1
    else:               # recursive case
        return n * factorial(n - 1)
```

#### How it works step-by-step for `factorial(3)`
```
factorial(3)
  → 3 * factorial(2)           # waiting for factorial(2)
        → 2 * factorial(1)     # waiting for factorial(1)
              → 1 * factorial(0)
                    → 0 == 0? → return 1   # base case reached
              → 1 * 1 = 1
        → 2 * 1 = 2
  → 3 * 2 = 6
```
The calls "stack up" until the base case is hit, then they resolve in reverse order, multiplying as they go.

### Key Takeaways
- Recursion breaks a problem into self-similar subproblems.
- Every recursive function needs a stopping condition (base case).
- Without a base case, you get infinite recursion (eventually a stack overflow).
- It’s especially natural for problems with a recursive structure (trees, sorting, divide-and-conquer, etc.).

[
  {
    "id": "chatcmpl-cf0a764075534f728f55b99dacf898ba",
    "object": "chat.completion.chunk",
    "created": 1781047647,
    "model": "accounts/fireworks/models/deepseek-v4-pro",
    "choices": [
      {
        "index": 0,
        "delta": {
          "role": "assistant"
        },
        "finish_reason": null,
        "raw_output": null
      }
    ],
    "usage": null
  },
  {
    "id": "chatcmpl-cf0a764075534f728f55b99dacf898ba",
    "object": "chat.completion.chunk",
    "created": 1781047647,
    "model": "accounts/fireworks/models/deepseek-v4-pro",
    "choices": [
      {
        "index": 0,
        "delta": {
          "reasoning_content": "We"
        },
        "finish_reason": null,
        "raw_output": null
      }
    ],
    "usage": null
  },
  "... 606 more chunks omitted ...",
  {
    "id": "chatcmpl-cf0a764075534f728f55b99dacf898ba",
    "object": "chat.completion.chunk",
    "created": 1781047647,
    "model": "accounts/fireworks/models/deepseek-v4-pro",
    "choices": [],
    "usage": {
      "prompt_tokens": 14,
      "total_tokens": 622,
      "completion_tokens": 608,
      "prompt_tokens_details": {
        "cached_tokens": 0
      }
    }
  }
]

Parameters

▶messages[]

arrayrequired

temperature

numberminimum: 0maximum: 2

max_tokens

numberexclusiveMinimum: 0

max_completion_tokens

numberexclusiveMinimum: 0

top_p

numberminimum: 0maximum: 1

frequency_penalty

numberminimum: -2maximum: 2

presence_penalty

numberminimum: -2maximum: 2

stream

boolean

▶stream_options{}

object

▶tools[]

array

tool_choice

response_format

▶modalities[]

array

▶audio{}

object

reasoning_effort

stringOptional reasoning control; availability and accepted values are model-dependent.

string

object

string

created

number

model

string

▶choices[]

array

▶usage{}

object

API Schemas (Raw)

Input

Output