As the cost of enterprise AI adoption reaches unsustainable levels, businesses are rapidly shifting their AI strategies from “token maxing” to “model maxing.”
Business Insider reported on July 5 (local time) that companies burdened by rising AI costs are moving away from indiscriminately relying on the latest frontier models and are instead optimizing model selection based on the specific needs of each task.
Model maxing refers to the practice of using different AI models according to a task's complexity. Companies now reserve premium, high-performance models such as Claude Fable 5 and GPT-5.5 for workloads that require advanced reasoning, including scientific research and complex agent orchestration. Simpler, repetitive tasks are increasingly being delegated to older models or lower-cost open-source alternatives to improve cost efficiency.
Brian Armstrong, CEO of Coinbase, recently reinforced this trend in a post on X. He predicted that while demand for AI will continue to grow explosively, within the next 12 to 18 months, around 80% of AI workloads will be handled by models that are 99% cheaper than today's leading systems. According to Armstrong, fewer than 20% of tasks will truly require the latest premium models.
The trend is already becoming evident across enterprises. AI startup Bold Metrics assigns different AI models and reasoning levels to individual teams based on the nature of their work. Instead of deploying the most expensive model company-wide, the firm optimizes AI usage by selecting the model best suited to each team's objectives.
This marks a notable shift from the industry's previous tendency to chase the newest AI models regardless of cost. Chris Marconi, co-founder of Hetchura, said many organizations avoid the more complicated process of identifying the optimal model for each task and instead simply follow the latest trend.
As model maxing gains momentum, demand is also growing for model routing solutions, which automatically direct enterprise requests to the most appropriate AI model. According to corporate spend management platform Ramp, the share of businesses using model routers has increased fivefold, rising from 1% last year to 5% this year.
David Gilmore of model routing company Rayline said many businesses adopt the latest AI models simply because they fear falling behind the technology curve. “Only after receiving massive API bills do they begin scaling back usage and focusing on efficiency,” he said, describing a pattern that has become increasingly common among enterprise customers.
