DeepSeek V4.1 Flash API

DeepSeek V4.1 Flash gives agent builders image input, tool calls and a long context in one model, from $0.135 per million input tokens.

Get API key Read the docs

1M token context
16B active parameters
Native image input
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-V4.1-Flash",
    messages=[
        {
            "role": "user",
            "content": "Why do sheep count people?"
        }
    ],
)

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

How to use the DeepSeek V4.1 Flash 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 pass deepseek-ai/DeepSeek-V4.1-Flash as the model string.
3
Send your messages to /chat/completions with reasoning_effort set as an integer, then read the reply from choices[0].message.content.

What DeepSeek V4.1 Flash is

DeepSeek V4.1 Flash is a mixture-of-experts model that reads images and text and generates text. DeepSeek released it in September 2026 as the smallest model in a new causal encoder-decoder family, built for faster inference and higher throughput. It holds 552B parameters and activates only 16B of them per token. It replaces V4 Flash and V4 Flash Vision Exp, which DeepSeek retired on the same day, and it scores 39.5 on the Artificial Analysis Intelligence Index.

What it does well

Drives a coding agent from an issue to a patch, and calls tools in a loop.
Takes an image and a question in one prompt and answers in text.
Holds a whole repository or a long contract in one prompt, up to 1M tokens.

What it is not for

It returns text only. It does not generate images or audio.
It writes long answers. Artificial Analysis measured its output volume well above the median of the models it tested, so output cost and run time climb on agent loops.

Calling it

Call it on POST /chat/completions with the model string deepseek-ai/DeepSeek-V4.1-Flash. Thinking stays on at middle effort when you send no setting, so pass reasoning_effort as an integer from 1 to 100. The names low, high and max map to different numbers on DeepSeek's own API than on the open weights, and thinking mode ignores temperature, presence_penalty and frequency_penalty.

deepseek-ai/DeepSeek-V4.1-Flash /chat/completions

SDKs and languages

DeepSeek V4.1 Flash 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, image
Returns	text
Total parameters	552B
Active parameters	16B
Score	39.5 on Artificial Analysis intelligence index
Languages	not published
Precision served	MXFP4/MXFP8
Context	1M tokens
Vector	n/a
Pulls	570.9K in the last 30 daysWeights pulled from Hugging Face, read on 2026-09-24. It counts the repository, not this API.
License	MIT

Price

Input

$0.135

per million tokens

Output

$0.540

per million tokens

Against OpenRouter

10%

below the cheapest of 19 providers

The cheapest OpenRouter price for this model was $0.150 in and $0.600 out, with a median of $0.300 and $1.200 across the comparable providers. Read on 2026-09-17.

License

DeepSeek publishes the weights and the repository under the MIT license. It permits commercial use, modification and redistribution, and it adds no extra vendor terms to read before you ship.

Parameters: 2.4T, 95B
Context: 1M
Input per 1M: $1.800
Output per 1M: $5.400
Released: Aug 2026
Pulls, 30 days: 54.4K
License: Qwen3.8 Max