Active

Dec 26, 2024

ProducerDeepSeek AI

FamilyDeepSeek

APIOpen WeightsLocal RunCloud

DeepSeek-V3

LLM

Open-weight Mixture-of-Experts language model with 671B total parameters (37B activated per token), developed by DeepSeek AI and released in December 2024.

Available on

Hugging Face Hub

Technical specification

Context window

Parameters

671B total, 37B activated

Max output

License

DeepSeek License v1.0

Tools

Yes

Fine-tuning

Yes

Weights access

Open weights

Knowledge cutoff

Jul 2024

Last updated: May 3, 2026

Modalities

Input

Text

Output

Text

Code

Structured data

summaries

reports

Capabilities

Reasoning★

Reasoning

Multi-step reasoning★

Reasoning

Long context★

Reasoning

Coding★

Coding

Function Calling

Planning

Structured output★

Structured gen.

Multilingual★

Language

Planning★

Planning

Streaming output

Reasoning

Architecture and technologies

Core Architecture

Training Techniques

Applications

Pricing

Apr 20, 2026Source

PublicUSDper 1M tokens

Prices refer to the current DeepSeek API (model deepseek-chat = DeepSeek-V3.2 via API from December 2025). Original DeepSeek-V3 prices at launch: input cache hit $0.07/1M, input cache miss $0.27/1M, output $1.10/1M (historical prices). The current API uses V3.2.

Pricing · Calculator

Token volume10K

INPUT

$0.2700 / per 1M tokens

$0.00270

OUTPUT

$1.1000 / per 1M tokens

$0.0110

CACHE

$0.0700 / per 1M tokens

$0.00070

TOTAL

for 10K tokens

$0.0137

All plans1

TextStandard/ per 1M tokens

Input$0.2700

Cache$0.0700

Output$1.1000

Historical pricing for DeepSeek-V3 (launched December 2024). Cache hit = $0.07/MTok, cache miss = $0.27/MTok, output = $1.10/MTok.

DeepSeek API obsługuje kontekstowe cachowanie (context caching). Wagi modelu dostępne bezpłatnie na Hugging Face i GitHub do samodzielnego wdrożenia. Ceny podane dla oryginalnej wersji DeepSeek-V3 (XII 2024); aktualne API deepseek-chat wskazuje na DeepSeek-V3.2.