DeepSeek-R1

DeepSeek-R1 is a 671B-parameter mixture-of-experts reasoning model (37B activated per token) released by DeepSeek in January 2025, trained with large-scale reinforcement learning to achieve benchmark-competitive performance on math, coding, and multi-step inference tasks. It is released as open weights and positioned as a cost-efficient alternative to closed frontier reasoning systems such as OpenAI o1. Source: supabase.