Does Every AI Task Need the Most Powerful Model?
31 October 2026
I read this interesting Reuters article: SAP swaps costly AI models for cheaper alternatives, targets over €2 billion in gains.
It reveals quite a bit about how enterprises are beginning to think about AI.
According to what’s written, SAP is increasingly replacing expensive AI models with cheaper alternatives wherever they can deliver sufficiently good results.
SAP reportedly uses more than 70 AI models from providers including OpenAI, Anthropic and Google. Rather than automatically choosing the most powerful model, the company is looking at which model can perform a particular task reliably at the lowest cost. It bears mentioning that SAP’s CFO, Dominik Asam, said the company expects overall AI spending to increase as usage grows. The goal is to keep costs under control as AI scales, rather than simply reduce spending.
We have seen similar thinking emerge in conversations with other organizations as well.
For a while, much of the AI conversation has been about which model is the smartest, which performs best on benchmarks, and which has the latest capabilities.
But once AI becomes part of everyday business workflows, the questions are beginning to change.
Do we really need the most capable model for every task? Or just one that gets the job done well enough, at a reasonable cost?
