Mollick: AI Labs Could Beat Open Weights on Cost-Performance

Ethan Mollick argues that labs like OpenAI and Anthropic could increasingly beat open-weight models on cost-performance by offering much cheaper “weaker” versions of frontier systems and compounding gains from inference scaling and R&D. He adds that some companies will still prefer open models for control and other practical reasons.

2026-07-07 ~ 2026-07-07 · 2 related posts