How much does DeepSeek V4 Pro cost?

The listed rate is $0.435 per million input tokens and $0.87 per million output tokens. Cache, long-context, media, gateway, or promotional pricing can differ; see the pricing note on this page.

What is DeepSeek V4 Pro's context window?

The provider advertises up to 1,000,000 input tokens and 384,000 output tokens. PromptHQ or individual gateway providers may impose lower operational limits.

Does DeepSeek V4 Pro support reasoning?

Yes. DeepSeek V4 Pro supports high and max reasoning settings through supported API routes. The response shown here came from DeepSeek's first-party chat interface, where the exact setting was not recorded.

Model guide / DeepSeek

DeepSeek V4 Pro

current

DeepSeek V4 Pro is an open-weight, one-million-context reasoning model aimed at coding, STEM, and agents. Its 1.6-trillion-parameter mixture-of-experts architecture activates 49 billion parameters per token and pairs frontier-oriented performance with unusually low official API prices, though it is text-only and the largest reasoning mode is token-intensive.

Released Apr 24, 2026MITOpen weightsVerified Jul 20, 2026

Try DeepSeek V4 Pro See our evaluation

Context window

1M tokens

Maximum output

384K tokens

Input price

$0.43 / 1M

Output price

$0.87 / 1M

Best for

Cost-sensitive reasoning and coding agents
Very long text and code contexts
Teams that need open weights or private deployment
OpenAI- or Anthropic-compatible API integrations

Not ideal for

Image, audio, or video input
Low-latency tasks where V4 Flash is sufficient
Workflows unable to budget for long reasoning outputs
Teams needing a fully mature, non-preview release

PromptHQ evaluation

PromptHQ 100-word challenge

We give every model the same miniature science-fiction mystery. The response is short enough to read in one chat bubble, while the premise tests creative writing, compression, foreshadowing, causal coherence, and instruction following.

DeepSeek V4 Pro102 words

Write a science-fiction mystery in exactly 100 words. A three-person research station on Europa receives a message from itself, timestamped 11 minutes in the future: "Do not open the greenhouse door." The door is opened anyway. Plant two subtle clues. Give the warning a logical explanation without aliens, time travel, hallucinations, or simulations. The final sentence must change what the reader thought the warning meant. Output only the story. Do not explain it.

The message came through the ag-console: origin Europa Station, timestamp +00:11. "Do not open the greenhouse door."

Yoshida saw the ethylene warning nested beneath — imminent fruiting spike. "System glitch," she said.

Daviau opened it anyway.

Inside, engineered wheat stood motionless under amber lamps, each stalk tipped with nascent grain. Ren crossed her arms. They waited. Nothing moved. Nothing happened.

Eleven minutes later, the wheat yellowed — every stalk at once, grain shriveling to black husks. The ag-system chimed: Oxygen spike detected. Premature senescence. Crop viability: zero. The warning had been for the wheat. Not for them. And they had killed it.

Run the same prompt yourself

Open DeepSeek V4 Pro in Multichat with the evaluation prompt prefilled.

Try this prompt

What the response revealed

Exact word countfail

The response contains 102 whitespace-delimited words, two over the required count.

Constraint followingpass

It outputs only the story and avoids aliens, time travel, hallucinations, and simulations.

Causal coherencefail

Opening the door plausibly causes the crop-damaging oxygen spike, but the response never explains how the station produced a message timestamped eleven minutes in the future.

Foreshadowingpass

The ethylene warning, imminent fruiting spike, and nascent grain plant multiple clues that the crop—not the crew—is at risk.

Endingmixed

The closing sequence reframes the warning as protection for the wheat, but the literal final sentence only confirms the destruction after the reinterpretation has already been stated.

Writingpass

The pacing is crisp and the synchronized crop failure creates a strong visual turn, despite the unresolved timestamp mechanism.

The prompt was submitted unchanged through DeepSeek's first-party chat interface. The API route, reasoning setting, token usage, and latency were not recorded. We preserve the response exactly as supplied apart from display rendering.

Performance

DeepSeek V4 Pro benchmarks

Benchmark scores are sensitive to reasoning effort, harness, tools, token budget, prompt format, sampling, and evaluation date. Scores here retain their source and should not be treated as directly interchangeable unless the underlying setup matches.

Benchmark	Category	Score	Reported by
SWE-Bench Verified	software engineering	80.6%	LLM Reference³
SWE-Bench Pro	software engineering	55.4%	LLM Reference³
LiveCodeBench	competitive programming	93.5%	LLM Reference³
GPQA Diamond	graduate science reasoning	90.1%	LLM Reference³
Terminal-Bench 2.0	terminal and agentic coding	59.1%	LLM Reference³

SWE-Bench Verified

80.6%

software engineering · LLM Reference³

SWE-Bench Pro

55.4%

software engineering · LLM Reference³

LiveCodeBench

93.5%

competitive programming · LLM Reference³

GPQA Diamond

90.1%

graduate science reasoning · LLM Reference³

Terminal-Bench 2.0

59.1%

terminal and agentic coding · LLM Reference³

Family position

DeepSeek V4 positioning

V4 Pro is the large, capability-focused DeepSeek V4 model; V4 Flash is smaller and faster. Pro is notable for open weights, a one-million-token context, and low first-party API prices.

DeepSeek V4 Pro

This model

Maximum V4 capability

$0.43 input

$0.87 output

DeepSeek V4 Flash

Fast V4 sibling

$0.07 input

$0.28 output

GLM 5.2

Open-weight coding rival

$0.93 input

$3 output

API pricing

Per million text tokens

Input

$0.43

Cached input

$0.00

Cache write

$0.43

Output

$0.87

DeepSeek's cache-hit price is much lower than cache-miss input. OpenRouter pricing can vary by host; this page shows DeepSeek's official API list rates.

Where DeepSeek V4 Pro stands out

Built for demanding work

Cost-sensitive reasoning and coding agents are central to the model's positioning, rather than an incidental capability.

Long-context capacity

The published context window is 1,000,000 tokens, making the model a candidate for large documents, repositories, and sustained agent state.

Reasoning and tools

Reasoning is supported with high, max provider setting(s), and the model can participate in tool-using workflows through its available API surface.

Limitations to know

Benchmarks are configuration-sensitive

Scores can move substantially with the harness, tool access, effort setting, token budget, and evaluator. Treat the table as evidence, not a universal ranking.

Context size is not guaranteed recall

A large advertised window does not mean every detail is retrieved reliably at maximum length. Validate representative long-context workloads before deployment.

Product access differs from model capability

PromptHQ and gateway limits may expose fewer modalities, tools, or tokens than the provider's first-party API.

Capabilities and specifications

Knowledge cutoff

Not publicly disclosed

Inputs

text

Reasoning

Supported

Default effort

high

Supported API features

Streaming

Function calling

Structured outputs

Tool search

Web and file search high effort

Frequently Asked Questions

Sources

1
DeepSeek API models and pricing
DeepSeek · official documentation
2
DeepSeek V4 Preview Release
DeepSeek · official launch post
3
DeepSeek V4 Pro model reference
LLM Reference · independent benchmark summary
4
DeepSeek V4 Pro API pricing and availability
OpenRouter · gateway model page
5
PromptHQ model registry
PromptHQ · internal product configuration

Compare leading AI assistants

ChatGPT vs Claude ChatGPT vs Gemini Claude vs Gemini

See how DeepSeek V4 Pro handles your own work.

Try DeepSeek V4 Pro in Multichat

Available on PromptHQ Plus, Max