DeepSeek V4

deepseek-v4

Open Source

About

DeepSeek V4 is the April 2026 DeepSeek model family for long-context reasoning and coding. It covers the flagship V4 Pro model and the lower-cost V4 Flash variant, pairing a 1M-token context window with hybrid sparse attention, tool use, structured outputs, and reasoning modes.

DeepSeek V4 has a 1M-token context window.

Capabilities

VisionMultimodalReasoningFunction CallingTool UseStructured OutputsCode ExecutionPrompt CachingBatch APIAudioFine-tuning

Rankings

Best AI Agents & Agentic Models Best Open Source LLMs Best LLMs for Reasoning & Math Best LLMs for Function Calling & Tool Use Cheapest LLM APIs Best LLM APIs Best LLMs for Enterprise Best Free LLMs Best LLMs for Writing Best LLMs for Marketing Best LLMs for Customer Support

Compare

All comparisons →

DeepSeek V4 vs Kimi K2.6 DeepSeek V4 vs Grok 4 DeepSeek V4 vs Claude Opus 4.7 DeepSeek V4 vs DeepSeek V3.1 DeepSeek V4 vs Gemini 3.1 Pro DeepSeek V4 vs GPT-5 DeepSeek V4 vs Qwen3.6-Max DeepSeek V4 vs Claude Sonnet 4.6 DeepSeek V4 vs GPT-5.2 DeepSeek V4 vs GPT-5.5 DeepSeek V4 vs Qwen3-Max DeepSeek V4 vs GPT-4o (11-20)DeepSeek V4 vs GPT-5.5-Cyber DeepSeek V4 vs DeepSeek R1 DeepSeek V4 vs Llama 4 Maverick 17B Instruct FP8 DeepSeek V4 vs Gemini 2.5 Flash Live API

Specifications

FamilyDeepSeek V4

Released2026-04-24

Parameters1.6T

Context1M

Max output384,000

ArchitectureMixture of Experts

Specializationgeneral

LicenseMIT

Trainingpretrained

Created by

Advancing artificial general intelligence (AGI).

Hangzhou, Zhejiang, China

Founded 2023

Links