| Mobile Web

AI model economics shift as cost recovery for cutting-edge models gets harder

The share of AI use accounted for by cost-effective models is rising as demand concentrates on cheaper systems with slightly lower performance rather than top-tier models. Venture capitalist Tomasz Tunguz highlighted data showing most tokens on OpenRouter are handled by non-top models that are far cheaper. He said gains among leading models and narrowing performance gaps, along with buyer price sensitivity and vendor lock-in, may make it increasingly difficult to recoup training costs that can run into hundreds of millions of dollars.