DeepSeek-V3 / R1
Open MoE + reasoning models that redefined cost efficiency
DeepSeek-V3 is a ~671B-parameter Mixture-of-Experts model, and R1 is its open reasoning model, both released with open weights and famously trained at a fraction of Western frontier cost under chip-export constraints. Their January 2025 release triggered a global reassessment of AI training economics and a Nvidia market-cap shock. They are a leading proof that efficiency can substitute for raw compute.
Maker
DeepSeek
Architecture
MoE (~671B) + R1 reasoning
License
Open weights
Trained on
Export-restricted H800/H20
How it fits the stack
DeepSeek-V3 / R1 with what it depends on (above) and what it feeds (below). The figure renders as a crawlable diagram and upgrades to an interactive 3D graph as it scrolls into view.
DeepSeek-V3 / R1 in the AI stack. DeepSeek-V3 / R1 with its immediate upstream dependencies (top) and downstream dependents (bottom) in the AI value chain. Hover a node in 3D, or read the full relationships below.
Graph data (text) — 7 entities, 6 relationships
- DeepSeek-V3 / R1 —hosts→ Hugging Face
- DeepSeek-V3 / R1 —uses→ SGLang
- DeepSeek-V3 / R1 —uses→ Synthetic data & distillation
- DeepSeek-V3 / R1 —uses→ Transformer / MoE Architecture
- DeepSeek R2 / V3.x —depends on→ DeepSeek-V3 / R1
- DeepSeek-V3 / R1 —designs→ DeepSeek
Depends on ↑ · 4
Feeds ↓ · 1
Context — capital, rivals, policy · · 1