Routers Are a New AI Primitive (www.vals.ai)

In a significant development for the AI/ML community, Vals has introduced routers as a new primitive designed to optimize the selection of AI models for various tasks across multiple domains. As multiple language models release rapidly, organizations face the challenge of choosing the best model for specific workloads, balancing performance with cost and complexity. The Vals Index, which evaluates models across rigorous benchmarks in areas like finance and legal research, highlights that no single model excels across all tasks. This variance necessitates a system that can dynamically select the most suitable model for each request, a gap that routers aim to fill. Routers operate by evaluating incoming requests and determining the best model based on criteria such as task difficulty, cost, and expected latency. This decision-making process occurs continuously throughout a task's lifecycle, rather than just at the outset, which is crucial for complex, multi-step agent workflows. By employing an "oracle router" model for comparison, Vals found that the right ensemble of models could outperform any single model by a notable margin while potentially reducing costs by about 40%. The ongoing evaluation and benchmarking of these routers against established models promise to refine their performance and validate their effectiveness in real-world applications, making them critical tools in leveraging AI advancements effectively.