DeepSeek V4 Pro API

Call the preview checkpoint of DeepSeek's large V4 model for long context reasoning, code and agent runs, from $0.939.

1M context window
1.6T total parameters
37B active parameters
MIT licensed weights

from openai import OpenAI

client = OpenAI(
    base_url="https://api.modelsheep.com",
    api_key=API_KEY_MODELSHEEP,
)

response = client.chat.completions.create(
    model="deepseek-ai/DeepSeek-V4-Pro",
    messages=[
        {
            "role": "user",
            "content": "Why do sheep count people?"
        }
    ],
)

print(response.choices[0].message.content)

How to use the DeepSeek V4 Pro API

1
Create a ModelSheep account and copy an API key from the dashboard.
2
Point the OpenAI SDK at the base url https://api.modelsheep.com and set the model to deepseek-ai/DeepSeek-V4-Pro, the undated preview checkpoint, not the 0813 build.
3
Call /chat/completions with your messages and leave reasoning_effort out, because it changes nothing on this build, and read the answer from the first choice of the response.

What DeepSeek V4 Pro is

DeepSeek V4 Pro is the preview checkpoint of the V4 Pro line. It is a mixture of experts model, so it holds 1.6T parameters but runs 37B on each token. DeepSeek built it for long context reasoning, code and agent work. It pairs Compressed Sparse Attention with Heavily Compressed Attention to cut the cost of a million token prompt. The dated build, DeepSeek V4 Pro 0813, is the official release that supersedes this preview. It keeps the same structure, adds a DSpark speculative decoding module and scores higher on the published benchmarks.

What it does well

Answers questions across a full repository in one prompt. DeepSeek reports MRCR 1M, CorpusQA 1M and LongBench-V2 at a million tokens.
Drives a multi step coding agent. DeepSeek claimed open source state of the art on agentic coding benchmarks at the preview.
Works through hard code and maths problems in Think Max, the highest of its three reasoning modes.

What it is not for

Production agent work. DeepSeek marks the 0813 build as the official release and reports stronger agent results there.
Images, video and audio. The model takes text and returns text.

Calling it

Post to /chat/completions with the model string deepseek-ai/DeepSeek-V4-Pro. ModelSheep serves the FP8 build. The model has three reasoning modes: Non-think, Think High, and Think Max. No API field selects them. The mode sits in the prompt encoding, and the preview repository ships no Jinja chat template.

deepseek-ai/DeepSeek-V4-Pro /chat/completions

SDKs and languages

DeepSeek V4 Pro answers on an HTTP endpoint, so any language that can post JSON reaches it. These are the ones the box above writes out.

OpenAI SDK: PythonTypeScriptGoJavaC#PHPRust
Anthropic SDK: PythonTypeScriptGoJavaC#PHPRust
HTTP: cURLPythonTypeScriptGoRustJavaC#PHP

Specifications

Author	DeepSeek
Category	language
Accepts	text
Returns	text
Total parameters	1.6T
Active parameters	37B
Score	30.9 on Artificial Analysis intelligence index
Languages	not published
Precision served	FP8
Context	1M tokens
Vector	n/a
License	MIT

Price

Input

$0.939

per million tokens

Output

$1.879

per million tokens

Against OpenRouter

10%

below the cheapest of 13 providers

The cheapest OpenRouter price for this model was $1.044 in and $2.088 out, with a median of $1.650 and $3.301 across the comparable providers. Read on 2026-09-17.

License

DeepSeek publishes the repository and the weights under the MIT license. The license permits commercial use. No extra vendor terms gate a deployment.

Similar models

Minimax M3MiniMaxAI/Minimax-M3

Parameters: 427B, 23B
Context: 1M
Input per 1M: $0.252
Output per 1M: $0.990
License: MiniMax Community

Qwen3.8 27BQwen/Qwen3.8-27B-FP8

Parameters: 27.8B
Context: 1M
Input per 1M: $0.180
Output per 1M: $1.980
License: Apache 2.0

Kimi K2.6moonshotai/Kimi-K2.6

Parameters: 1T, 32B
Context: 262K
Input per 1M: $0.513
Output per 1M: $2.160
License: Modified MIT

Start using DeepSeek V4 Pro

Point the OpenAI SDK at api.modelsheep.com. Your prompts stay in Germany, and nothing is retained.

Get API key Read the quickstart