Model routing squeezes OpenAI and Anthropic revenue

CNBC reports that corporate customers are shifting toward model routing, a practice that matches each task to the most cost-effective model rather than defaulting to the largest frontier system. CNBC attributes to Glean CEO Arvind Jain an estimate that roughly 95 percent of enterprise AI usage still runs on the most expensive frontier models. CNBC also cites Cognition CEO Scott Wu, whose company makes the coding agent Devin, saying firms can achieve about five to ten times better cost efficiency on routine work by using smaller models. CNBC ties the trend to CFOs and boards cracking down on AI budgets that have run ahead of expectations.

Routing reduces average cost per query by assigning low-complexity requests to cheaper, lower-latency models and reserving frontier models for hard problems. As a general engineering pattern, teams implement it with lightweight classifiers, heuristics, or scorers plus instrumentation to track latency and cost. If routing spreads, the volume of queries flowing to premium-priced models could fall, which CNBC notes would weaken revenue pools for providers that price primarily for top-capability usage and that have built IPO expectations on sustained premium demand.

• •Adoption signals: enterprise case studies, vendor disclosures, or telemetry showing shifts in query mix.
• •Tooling: open-source or managed services that simplify routing and model orchestration.
• •Pricing: new tiers or contract terms from providers that reflect routing-aware consumption.

For practitioners, the takeaway is to benchmark models against realistic workload mixes and total cost of ownership, and to build reliable routing logic and monitoring so cheaper models handling production traffic do not cause quality regressions.