AI model costs surge as companies default to flagship versions

Companies are facing significant rising costs due to the inefficient use of high-end artificial intelligence models. While flagship models from providers like Anthropic and OpenAI offer superior performance, they can cost up to 100 times more per token than smaller, more efficient versions.

Major technology firms are experiencing substantial expenditures. Meta employees reportedly consumed 60 trillion AI tokens in a single month, an amount estimated to cost roughly $900 million at Anthropic’s list prices. Similarly, Uber reportedly exhausted its 2026 AI tools budget within just four months.

Experts suggest that many organizations default to using the most expensive models for simple tasks, such as generating emails, rather than routing specific tasks to more cost-effective models. Strategies to mitigate these costs include trimming context, caching, batching, and implementing smarter model routing.