Skip to contentATP Token

DeepSeek V4 Pro

DeepSeekAll DeepSeek models

Input

$2.40

USD / 1M tokens

Output

$4.80

USD / 1M tokens

Context

1M

tokens

Released

2026-04-24

Positioning: Flagship. Price level 2 of 4 among text models.

DeepSeek V4 Pro is the flagship of DeepSeek's V4 family, a mixture-of-experts model with 1.6T total and 49B active parameters per DeepSeek. Its GA build, DeepSeek-V4-Pro-0813, arrived in August 2026 with stronger agent performance, thinking on by default, and low/high/max effort levels. DeepSeek says it will keep serving V4 Pro on its API alongside the newer V4.1-Flash.

ThinkingToolsVision
Get started freeAll model pricesCompare

DeepSeek's Models & Pricing page lists a 1M context length and a maximum output of 384K tokens for deepseek-v4-pro.

Pricing

Pay for what you use: credits are deducted only when a request succeeds.
Prices in USD · $1 = 100 credits
Input
$2.40
Cache read
$0.20
Output
$4.80
USD / 1M tokens

DeepSeek · Family · USD / 1M tokens

Text models from the same vendor on ATP. Bars fill up to the family's highest price.
All DeepSeek models
ModelInputOutput
DeepSeek V4 Pro This pagedeepseek-v4-pro$2.40$4.80
DeepSeek V3.2deepseek-v3-2$0.57$1.71
DeepSeek V4 Flashdeepseek-v4-flash$0.20$0.40

Cost per 1,000 typical requests

A quick estimate with fixed token counts per request. No caching, no retries. Try the estimator with your own numbers.
WorkloadUSD / 1,000
Short chat turn800 / 400$3.84
RAG answer4,000 / 500$12.00
Agent step with tools20,000 / 1,500$55.20

Where it fits

Good fit
  • Agentic coding and terminal workflows that benefit from extended thinking
  • Long-horizon, multi-tool agents where effort can be tuned per task
  • Synthesis across large codebases or document sets in a 1M-token context
  • Math, STEM, and analytical reasoning tasks
Usually not ideal
  • Bulk, latency-first traffic — deepseek-v4-flash on ATP is the lighter tier
  • Compliance scopes that mandate OpenAI- or Anthropic-only models

Benchmarks

Vendor-reported
  • Terminal Bench 2.187.9

    DeepSeek-V4-Pro-0813 (GA); Code Agent tasks: DeepSeek Harness minimal mode, max effort, top_p=0.95, temperature=1.0 (per DeepSeek change-log notes)

    DeepSeek
  • HLE (wo / w tools)42.7 / 60.0

    DeepSeek-V4-Pro-0813 (GA); without tools / with tools

    DeepSeek
  • DeepSWE62.7

    DeepSeek-V4-Pro-0813 (GA); Code Agent tasks: DeepSeek Harness minimal mode, max effort, top_p=0.95, temperature=1.0 (per DeepSeek change-log notes)

    DeepSeek
  • NL2Repo61.5

    DeepSeek-V4-Pro-0813 (GA); Code Agent tasks: DeepSeek Harness minimal mode, max effort, top_p=0.95, temperature=1.0 (per DeepSeek change-log notes)

    DeepSeek
  • Toolathlon-Verified74.1
    DeepSeek

Scores as published by each vendor, not measured by ATP. Test conditions vary, so they are not directly comparable across vendors.

Same controls on every model: project keys, per-project model access, credit budgets, and a log of every request.Set up your organization

Cost estimator

Plug in your usage to see roughly what a month costs.
0200
0100
ESTIMATE
$33.60
USD / month
3,360 credits / month
  • Input$24.00
  • Output$9.60

$33.60 USD / month · 3,360 credits / month

An estimate at our USD prices (USD 1 = 100 credits). What each request actually cost shows up in your console's request log. Cached input is billed at the lower cache-read price, so you may spend less.

Quickstart

Keep your SDK. Point it at ATP and use a project key. Read the docs
quickstart.py · OpenAI SDK
from openai import OpenAI

client = OpenAI(
  base_url="https://api.atptoken.ai/v1",
  api_key="atp-•••"
)

r = client.chat.completions.create(
  model="deepseek-v4-pro",
  messages=[{"role":"user","content":"hi"}],
)

Compare with similar models

Same type and positioning, from other vendors, closest in price first.
Compare side by side

Related models

All DeepSeek models
DeepSeek V3.2deepseek-v3-2$0.57 / $1.71In / out · per 1M tokensDeepSeek V4 Flashdeepseek-v4-flash$0.20 / $0.40In / out · per 1M tokensGPT-5.4 Progpt-5.4-pro$30.00 / $180.00In / out · per 1M tokensClaude Fable 5claude-fable-5$10.00 / $50.00In / out · per 1M tokensGPT-5.6 Solgpt-5.6-sol$5.00 / $30.00In / out · per 1M tokensGPT-5.5gpt-5.5$5.00 / $30.00In / out · per 1M tokens

FAQ

4
Can DeepSeek V4 Pro use tools, read images, or think before answering?
DeepSeek V4 Pro supports tool calling, vision input and extended thinking, but not audio.
How much context does DeepSeek V4 Pro take?
DeepSeek lists a 1,000,000-token context window for DeepSeek V4 Pro. Source: DeepSeek API Docs — Models & Pricing. The model list your key can see shows the limit you actually get.
Is cached input cheaper on DeepSeek V4 Pro?
You pay per 1M input tokens and per 1M output tokens. Cached input costs 8% of the normal input price. You only pay when a request succeeds. $1 = 100 credits. See pricing
How do I call DeepSeek V4 Pro on ATP?
Create a project API key and point any OpenAI-compatible SDK at https://api.atptoken.ai/v1. Set model to "deepseek-v4-pro" (the API id, not the display name). Make sure the model is turned on for your project.

Sources

The model description on this page cites these sources. Prices come from ATP.
Get started

Start with $5. Credits never expire.

Create a project key, turn this model on, and pay only for what you use.

Get started freeTalk to enterprise