DeepSeek
A leading AI research company focused on developing advanced language models and AI technology for coding and general intelligence.
Models
DeepSeek V3.2
DeepSeek V3.2 is a reasoning-first model with 128K context window, designed for agent tasks with integrated thinking in tool-use capabilities.
DeepSeek-Coder-V2 Instruct 0724
Open-weight MoE coding model supporting 338 programming languages, code completion, insertion, tools, and JSON output.
DeepSeek-R1-0528
Updated DeepSeek-R1 reasoning checkpoint with improved coding, mathematics, and reduced hallucination behavior.
DeepSeek R1
DeepSeek's open-weight 671B MoE reasoning model for mathematics, coding, and complex problem solving.
DeepSeek-V3.1
Hybrid thinking and non-thinking model with stronger agentic coding, tool use, and a 128K context window.
DeepSeek-V3.2-Exp
Experimental V3.2 checkpoint introducing sparse attention for more efficient long-context coding and agent tasks.
DeepSeek V3.1 Terminus
DeepSeek V3.1 Terminus improves language consistency and optimizes code-agent and search-agent behavior over V3.1.
DeepSeek-V3
Open-weight MoE model that established the V3 coding, reasoning, and general-purpose model line.
DeepSeek-V4-Flash
DeepSeek's efficient open-weight V4 model with 284B total and 13B active parameters, 1M context, dual thinking modes, JSON output, and tool calls.
DeepSeek-V4-Pro
DeepSeek's flagship open-weight V4 model with 1.6T total and 49B active parameters, 1M context, dual thinking modes, JSON output, and tool calls.
Community
Follow us
X (Twitter)
Follow us
GitHub
View repos
YouTube
Watch videos
Discord
Join community
Join discussion
Blog
Read blog posts