DeepSeek R1
by DeepSeekopen weightsDeepSeek R1 family
Parameters
671B
Context window
64K tokens
Released
2025-01-20
Input price / 1M tok
$0.55
Output price / 1M tok
$2.19
Modality
text
About DeepSeek R1
DeepSeek's open reasoning model — an open-weight competitor to OpenAI's o1. Trained with pure reinforcement learning on math/code tasks. Full reasoning trace visible in outputs. Sparked the 2025 conversation about efficient reasoning training.
Strengths
- Open weights
- o1-competitive reasoning benchmarks
- Cheap
- Visible chain-of-thought
Weaknesses
- No tools
- Verbose reasoning eats output tokens
- Text-only
Best for
MathResearchHard debugging