DeepSeek-V3
by DeepSeekHighly efficient 671B MoE model trained on 14.8T tokens. Achieves top benchmark scores at fraction of typical training cost.
Specifications
- Context Window
- 128,000 tokens
- Released
- December 2024
Pricing
- Input
- $0.27/MTok
- Output
- $1.10/MTok
Capabilities
MoE architectureEfficient trainingCodingMathOpen source