AT&T slashes AI costs 56% with model routers

AT&T's reported cost savings arrive as businesses grapple with rising AI expenses, a trend highlighted in June reports about companies seeking better cost management. The shift from chatbots to more compute-intensive agents, along with AI labs moving from flat subscriptions to token-based billing, has driven costs upward. This context helps explain why a large enterprise would invest in routing technology rather than simply cutting AI usage.

The company's strategy pairs cost-cutting routers with a deliberate push toward open-source models, whose capabilities generally trail frontier models by six to 10 months. Austin's observation that this gap is narrowing suggests AT&T sees open-source as an increasingly viable alternative to premium Anthropic and OpenAI models for many internal tasks. The evaluation of DeepSeek and Moonshot models indicates interest in expanding options further, though risk assessment remains a hurdle.

The reported end of "tokenmaxxing" — pushing employees toward the biggest models and heaviest usage — aligns with AT&T's approach. By matching task complexity to model capability, the company appears to be institutionalizing a more selective, cost-conscious AI strategy that other firms may study.