DeepSeek-R1
PermissiveDeepSeek · DeepSeek · Released 2025-01
671B total, 37B active (MoE)Text
The reasoning-focused sibling of DeepSeek-V3, released under a fully permissive MIT license, that triggered a wave of open reasoning-model distillation.
Strengths
- +MIT license, unrestricted use including distillation
- +State-of-the-art open reasoning benchmark results at release
- +Spawned widely-used distilled variants down to 1.5B
Limitations
- -671B total parameters require a multi-node GPU cluster to serve at full precision
- -Long chain-of-thought output increases latency and token cost
License
MIT
use commercially with attribution niceties
Hardware
needs multi-GPU or datacenter hardware to run well
Links
Stats
deepseek-ai/DeepSeek-R1— downloads·— likes
via Hugging Face