Model Family
DeepSeek
Chinese lab behind DeepSeek-V3 and R1, frontier-quality open MoE models at low cost.
Definition
DeepSeek's V3 and reasoning model R1 set new price-performance points for open-weights LLMs. V3 is a 671B-parameter MoE with 37B active per token. R1 trained with pure reinforcement learning rivals OpenAI's o1 on math benchmarks at a fraction of the inference cost.
Common use cases
- Cheap reasoning
- Self-hosting
- Open research