Skip to main content

​
Overview

deepseek-v4-pro is DeepSeek’s V4 flagship (April 2026). 1.6T total parameters with 49B active, MoE architecture, MIT license. It targets the top reasoning tier — complex coding, multi-step analysis, and long-context tasks where quality matters more than speed.

​
Specs

FieldValue
Model IDdeepseek-v4-pro
Best forTop reasoning, complex coding, long context
Context window1,000,000 tokens
Max output tokens65,536
Input modalitiesText
Output modalitiesText
Tool callingYes
Structured outputsYes
ReasoningYes
Prompt cachingYes
SpeedMedium
Cost bandPremium
Release stagePreview

​
Pricing

Per 1M tokens
Input$2.349
Output$4.698

​
Use this when

  • You’re doing serious reasoning or research where quality wins over latency.
  • You’re working with very long documents or repos and need 1M context.
  • You want the strongest open-weight DeepSeek tier available on Kyma.
  • You build coding agents that benefit from native tool calling and reasoning.

​
Pick something else when

  • You want a cheaper V4 option with the same family behavior: use deepseek-v4-flash.
  • You want the strongest agentic flagship: use kimi-k2.6.
  • You want the best general default at lower cost: use qwen-3.6-plus.
  • You only need the previous-gen DeepSeek flagship: use deepseek-v3.

​
Example

from openai import OpenAI

client = OpenAI(base_url="https://kymaapi.com/v1", api_key="ky-...")

response = client.chat.completions.create(
    model="deepseek-v4-pro",
    messages=[{"role": "user", "content": "Walk through the proof, then suggest a tighter bound and justify each step."}]
)