Currently, generating a usable-level code token requires consuming 100 to 1000 thinking tokens. This cost is plummeting annually, with a decline of over 10 times.

Even more interesting is that the ratio of thinking tokens to production-level tokens will soon collapse. What does this mean? Within two years, most code can be generated directly with a single or few prompts.

In plain terms, the cost structure of AI code generation is being completely rewritten. From burning money to usability, this cycle is rapidly shrinking.