DeepSeek-R1

model
DeepSeek
Incentivizing reasoning capability in LLMs via reinforcement learning. R1-Lite-Preview released 2024-11-20. Full R1 paper 2025-01-20. R1-0528 update released 2025-05-28.
Paper (arXiv)R1-Lite-Preview (2024-11-20)R1-0528 updateHuggingFaceArtificial Analysis

Model Details

Architecture MOE
Parameters 671B
Active params 37B
AA Intelligence 13 was 14 on v4.2

Variants

Name Parameters Notes
DeepSeek-R1-Lite-Preview — Released 2024-11-20
DeepSeek-R1 — —
DeepSeek-R1-0528 — Released 2025-05-28

Paper

arXiv HTML
reasoningopen-weighttraining