Robots Atlas>ROBOTS ATLAS
DeepSeek-V4
AI Modelsโ€บDeepSeek

DeepSeek-V4

V4ย ยทย Family: DeepSeek
Fourth-generation DeepSeek language models (V4-Flash and V4-Pro variants). MoE architecture, 1M-token context window. Preview: April 2026.
โœ“ Activeโœ“ Public accessLLMReasoning model๐Ÿ“ DeepSeek
Context window
1M
tokens
Parameters
284B (V4-Flash) โ€“ 1,6T (V4-Pro)
parameters
Release date
24 April 2026
๐ŸขDeepSeek AIProducer
Access:APIDownloadHostedDeployment:โ˜ Cloud๐Ÿ’ป Local

Overview

DeepSeek-V4 is the fourth-generation family of large language models from the Chinese company DeepSeek AI. The generation was released in preview on 24 April 2026 and ships in two variants: V4-Flash (about 284B parameters, a Mixture-of-Experts model under the MIT License) and V4-Pro (about 1.6T parameters, a commercial variant with closed weights). V4 models provide a 1M-token context window and use a Mixture-of-Experts (MoE) architecture built on the Transformer. Official releases: V4-Flash on 31 July 2026, V4-Pro on 13 August 2026.

Classification
LLMReasoning model
Family: DeepSeek
Access & deployment
APIDownloadHosted
CloudLocal
Key parameters
๐Ÿ“ Context: 1M
๐Ÿงฉ Parameters: 284B (V4-Flash) โ€“ 1,6T (V4-Pro)

Technical specification

Context window
1M
tokens
Parameters
284B (V4-Flash) โ€“ 1,6T (V4-Pro)
parameters
License
MIT (V4-Flash) / proprietary (V4-Pro)

Capabilities and applications

Native model capabilities
Advanced reasoning
The ability to perform multi-step, structured reasoning: analysing problems, planning steps, and drawing conclusions from hypotheses. Reasoning-first models (e.g. GPT-5.1 Thinking) dedicate a portion of inference to chains of thought before responding.
Category: reasoning
Coding
Generating, analysing and modifying code in many programming languages. Covers writing functions, debugging, refactoring, code review, and creating tests. Measured by benchmarks such as HumanEval and SWE-bench.
Category: coding
Long context
Support for large context windows โ€” tens to hundreds of thousands (or millions) of input tokens. Enables analysis of entire codebases, long documents, and many parallel conversations without losing earlier information. GPT-5.1 supports 400,000 tokens.
Category: language
Tool use
The model's ability to call external functions, APIs and tools during a conversation: calculator, search engine, code editor, database. The model decides when and how to use a tool and interprets its result.
Category: planning
Mathematical reasoning
The model's ability to solve mathematical tasks requiring multi-step reasoning โ€” equations, proofs, combinatorics, geometry, calculus and competition-level problems.
Category: reasoning

Technical architecture

Core Architecture
MOMoETRTransformer

Sources and related pages