Cost pressure is pushing companies to test smaller, cheaper AI models instead of always using the most advanced option.

Legal AI tool Harvey reduced inference costs by 3x without reducing quality in a test partnering with inference platform Fireworks AI, combining Claude Opus and Fireworks' GLM 5.1 and shifting to Opus for the most intensive tasks.

The shift reflects a change in how 'quality' is defined: instead of defaulting to the most powerful model for every task, companies are moving toward using the best model that gets the right answer most efficiently.

Coinbase co-founder Brian Armstrong predicted that 80% of workloads will run on 99% cheaper models within 12–18 months, with 20% of workloads still requiring latest generation models where maximum capability is important.