ML//model//open source model//DeepSeek

Chinese AI lab. V1 through V3 pushed open-weight models to frontier performance at a fraction of the cost.

Chinese AI lab. V1 through V3 pushed open-weight models to frontier performance at a fraction of the cost.

V3: Mixture of Experts architecture — 671B total params, ~37B active per token.

R1 (Jan 2025): matched o1-level reasoning using reinforcement learning with GRPO — no supervised fine-tuning for the reasoning phase.

Published full training recipes, open-sourced weights. Forced pricing collapse across the industry.