DeepSeek-R1
Open-weights reasoning model developed using large-scale reinforcement learning without cold-start supervised fine-tuning.
Best for: Self-hosted enterprise reasoning, local offline math/coding workloads, and sovereign AI deployments.
Specifications
- Provider: DeepSeek
- Release date: January 2025
- Context window: 128,000 tokens
- Modalities: Text, Code
- License: MIT Open Weights
- Pricing: $0.55 / MTok input, $2.19 / MTok output (or free self-hosted)
- API available: Yes
Capabilities
- Incentivized reasoning RL
- Full open weights and distillation
- Exceptional math and coding
- Low training cost parity