The cost of intelligence just hit zero
In November 2021, GPT-3 Davinci cost $60 per million tokens. By April 2026, DeepSeek delivers comparable performance at $0.14. The million-call application that cost $60,000 to run now costs roughly a hundred dollars. That descent is permanent because the floor is set by open-weight models that no single company can raise. Intelligence is commoditizing on the same curve as bandwidth and storage, and the value is migrating up the stack.