DeepSeek
Chinese model lab known for cost-efficient training narratives and open-weight drops that moved global price expectations.
DeepSeek’s 2025–2026 arc is a case study in how a relatively small team can shift market beliefs about inference economics—if the training and deployment story withstands replication. Western coverage oscillates between dismissal and panic; our default stance is to log primary sources, note hardware uncertainty, and separate price effects from capability claims.
The hub links DeepSeek’s reasoning and base-model lines to competitive responses from U.S. labs and cloud vendors.
Tracked claims attributed
Models shipped
- DeepSeek R1 — *MIT-licensed reasoning model associated with RL-style training narratives and sharp inference-price attention.*
- DeepSeek V3 — *Base-model generation preceding R1, cited in pricing shocks and replication attempts on MoE architectures.*