Description
DeepSeek AI is a Chinese AI research lab (affiliated with High-Flyer Capital Management) that became internationally prominent in 2024-2025 for releasing highly capable open-weight language models. Known for efficient training methods, mathematical reasoning research, and open-sourcing competitive frontier models at a fraction of reported Western training costs.
Key Properties
- Founded: 2023
- Location: Hangzhou, China
- Affiliation: High-Flyer Capital Management
- Focus: Frontier language model research, efficient training, mathematical reasoning
- Notable models: DeepSeek-V2, DeepSeek-V3, DeepSeek-R1, DeepSeek-Math
Notable Sources
- grpo-deepseekmath-group-relative-policy-optimization — DeepSeekMath (2024); introduced GRPO, pushed 7B math models to near-GPT-4 performance
- deepseek-r1-reasoning-via-reinforcement-learning — DeepSeek-R1 (2025); RL-trained reasoning model matching OpenAI o1 on math and coding benchmarks; fully open-sourced weights and training recipe
Related Entities
- openai — DeepSeek-R1 matched OpenAI o1 on several benchmarks