The Unbearable Cheapness of Open Weight Models
Summary
The article argues that open weight AI models are unusually cheap relative to frontier models, using DeepSeek as a focal point and comparing pricing to Anthropic and OpenAI. It raises questions about whether low costs are due to open-weight architecture, testability, or loss-leading strategies, and discusses broader implications for pricing, scarcity, and competition in AI. The piece also contrasts true open source pipelines with open weight offerings and references industry developments and partnerships fueling open approaches.