DeepSeek V4 Pro

DeepSeek V4 Pro (deepseek/deepseek-v4-pro) is DeepSeek’s flagship for complex tasks. It reasons at a controllable depth — high and xhigh — and can switch thinking off entirely when a task does not need it. Repeating context is read from cache at 9.7 ₽ per 1M tokens against the 116.1 ₽ full input price.

Parameters and capabilities

Parameter	Value
model	deepseek/deepseek-v4-pro (or the short deepseek-v4-pro)
Context	not published in the catalog
Reasoning	adjustable, can be disabled (off)
reasoning_effort	high, xhigh
Image input	no — an image_url part returns 400
File input	no — a file part returns 400
Cache read	9.7 ₽ per 1M tokens
Cache write	not published in the catalog

The effort set is specific: high and xhigh, without low or medium — you cannot request a depth below high. Values outside the list never reach the upstream: a non-existent level is silently replaced with the nearest one not deeper than requested, while off genuinely disables reasoning.

The catalog also carries the dated variant deepseek/deepseek-v4-pro-0813 with a different effort set (low/high/max) and different prices — check GET /v1/models when the exact version matters.

Examples

A complex task

Python

from openai import OpenAI

client = OpenAI(
    base_url="https://api.mixen.ai/v1",
    api_key="mxn-...",
)

resp = client.chat.completions.create(
    model="deepseek/deepseek-v4-pro",
    messages=[{
        "role": "user",
        "content": "Design a database schema for a marketplace: sellers, products, orders, reviews. Explain the trade-offs",
    }],
)
print(resp.choices[0].message.content)

cURL

curl https://api.mixen.ai/v1/chat/completions \
  -H "Authorization: Bearer $MIXEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek/deepseek-v4-pro",
    "messages": [{
      "role": "user",
      "content": "Design a database schema for a marketplace: sellers, products, orders, reviews. Explain the trade-offs"
    }]
  }'

Maximum depth and switching off

xhigh is the model’s deepest step; off removes reasoning where it is not needed:

curl https://api.mixen.ai/v1/chat/completions \
  -H "Authorization: Bearer $MIXEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek/deepseek-v4-pro",
    "reasoning_effort": "xhigh",
    "messages": [{"role": "user", "content": "Find the bottleneck in this algorithm and propose an asymptotically better one: <code>"}]
  }'

Price

Per 1M tokens: input 116.1 ₽ ($1.302), output 232.3 ₽ ($2.606), cache read 9.7 ₽ ($0.109); a separate cache-write price is not published by the catalog. Current prices — catalog.

Strengths and limits

• DeepSeek’s flagship tier — heavy tasks that need reasoning depth.
• Adjustable depth: high and xhigh, plus an honest off — simple tasks do not pay for thinking.
• Cheap cache reads — long sessions with repeating context get noticeably cheaper.
• No low or medium levels: the minimum controllable depth is high.
• Text only: images and files are not accepted as input — a request with them returns 400.
• Reasoning eats into max_tokens: on xhigh set a generous answer ceiling, or you will get finish_reason: "length".

All models — in the catalog. The general text-model workflow — chat guide.