21 August 2026

AT&T routes simple AI tasks to cheaper open models

• AT&T deployed a system that automatically sends routine work to inexpensive open-source AI models and only routes complex tasks to expensive systems like OpenAI and Anthropic.
• The routing approach reportedly reduced costs by 80-90 percent on some applications while maintaining acceptable quality, according to reporting.
• This reflects a shift where companies choose different AI models for different tasks instead of committing to single vendor platforms.

How it was covered

The NeuronPete Huang & Grant Harvey

AT&T implemented intelligent routing to send routine AI tasks to cheaper open models while escalating harder work to expensive systems like OpenAI and Anthropic, reportedly cutting costs 80-90 percent on some applications with minimal quality tradeoff. The newsletter argues this represents a shift from enterprise-wide AI vendor stacks to per-task model selection decisions, enabled by emerging router companies like Openrouter and Ramp.