Modelpeers
DeepSeek-V4-Pro character

ModelDeepSeek · Joined Apr 24, 2026

DeepSeek-V4-Pro

Flagship Model at DeepSeek | Open-weight reasoning and agentic coding | 1M-token context

Available for Agentic coding and math and reasoning

Compare
Official pageReport a correction

The basics

Checked Oct 22
Status
AvailableGenerally available since Aug 13, 2026. The API serves DeepSeek-V4-Pro-0813.
Model ID
Price
$1.32 in, $3.96 outPer 1M tokens, changed Aug 16
Context
1M tokens
Speed
42.3 tokens a second30-day median, first token in 0.33s

Updates

Security

DHS orders US federal civilian agencies to remove DeepSeek products

A Department of Homeland Security order published in the Federal Register on October 20, 2026 tells federal civilian agencies to stop buying DeepSeek products, including third-party products built with DeepSeek technology, and remove them from their systems. Agencies can request exceptions within 60 days.

Action needed:
If you sell to US federal civilian agencies, check whether your product uses DeepSeek models.

Default change

DeepSeek keeps V4-Pro running after September 14

DeepSeek had said deepseek-v4-pro requests would move to V4.1-Flash from September 14. It now keeps serving V4-Pro after that date at the same prices: $1.32 input (cache miss) and $3.96 output per 1M tokens at peak.

Before

deepseek-v4-pro routed to V4.1-Flash from September 14

After

deepseek-v4-pro keeps serving DeepSeek-V4-Pro-0813

Full changelog, by month

Strengths and weaknesses

Strengths

  • Agentic coding

  • Math and reasoning

  • Competitive programming

  • Long-context retrieval

Weaknesses

  • !

    Peak-hour prices are double the off-peak rates

  • !

    US federal civilian agencies were ordered to remove DeepSeek products

  • !

    Text only, with no image input

  • !

    DeepSeek announced a phase-out in September, then kept the model

Role

Open-weight model at DeepSeek

  1. DeepSeek-V4-Pro character

    DeepSeek-V4-ProCurrent

    Apr 2026 to now · 7 months

Works with

From recorded changes
DeepSeek API

API · 3 shared changes

Perplexity Agent API

API · 1 shared change

Similar models

DeepSeek-V4.1-Flash character
DeepSeek-V4.1-Flash

DeepSeek's smallest model in its new architecture family, a 552B-parameter MoE with native visual understanding.

Kimi K3 character
Kimi K3

$3 in, $15 out

GLM-5.3 character
GLM-5.3

$1.40 in, $4.40 out

Similar models for now. Real viewing data comes after launch.

Cost vs peers

2nd cheapest of 5 similar models by output price.

Standard API rates for uncached input on the shortest prompts, from each maker's pricing page, or from models.dev for models not researched here yet. DeepSeek-V4-Pro: Peak hours. Off-peak prices are half.

Cost and specs

Cost

Per 1M tokens

Input, cache miss (peak)
$1.32
Output (peak)
$3.96
Input, cache hit (peak)
$0.044
Input, cache miss (off-peak)
$0.66
Output (off-peak)
$1.98
Input, cache hit (off-peak)
$0.022

Specs

Context
1M tokens
Max output
384K tokens
Parameters
1.6T total, 49B active
Version served
DeepSeek-V4-Pro-0813
Modes
Thinking (default) and non-thinking
Reasoning effort
Low, high and max
Features
JSON output, tool calls and fill-in-the-middle (non-thinking only)
Input
Text
License
MIT

Where it works

PlacePrice per 1M tokens, in / outUptime
DeepSeek APIdeepseek-v4-pro
chat.deepseek.com and DeepSeek appExpert Mode
OpenRouterdeepseek/deepseek-v4-pro-0813
Vercel AI Gatewaydeepseek/deepseek-v4-pro-0813
Microsoft Azuredeepseek-v4-pro$1.91 / $3.83100%
Self-hostedOpen weights on Hugging Face, MIT license
Relace$0.90 / $1.50100%
StreamLake$0.80 / $1.60100%

Prices and uptime are OpenRouter's figures for each provider's cheapest endpoint. Uptime covers the last 24 hours.

Benchmarks

Higher is better. Green is DeepSeek-V4-Pro. Dark marks are similar models.

Independent

Self-reported by DeepSeek

SWE-bench Verified

Real GitHub issues

80.6

Codeforces

Competitive programming rating

3206

BrowseComp

Hard web browsing questions

83.4

What people say

Reviews

No reviews yet. Be the first to review DeepSeek-V4-Pro.

Used DeepSeek-V4-Pro? Tell other builders what worked and what did not.

Incidents

Writing habits

Creative writing · EQ-Bench, among 51 models tracked here

Slop score

19.7

Overused AI words and phrases. Less slop than 32% of tracked models.

Repetition

3.2

Repeated words and ideas. Less repetition than 82% of tracked models.

Story length

7,444

Average characters per story. Longer than 72% of tracked models.

Research sources

Researched Oct 22, 2026. Each section's own sources open with its Show sources button.