Show HN: Tier – Adaptive tool routing that makes small LLMs 10pt more accurate (github.com)

A new tool called "Tier" has been introduced, leveraging adaptive tool routing to enhance the performance of small language models (LLMs), increasing their accuracy by up to 10 percentage points. Traditional AI agent frameworks present identical tool options to models of all sizes, overwhelming smaller models with irrelevant choices. Tier addresses this challenge by customizing tool visibility based on the model's size, optimizing navigation rather than selection. For instance, a 1.5B model now sees only the most relevant tools with detailed descriptions, leading to significant improvements in performance and token efficiency. This development is particularly significant for the AI/ML community as it enables better utilization of smaller LLMs, which are often constrained by computing resources. The Tier system employs a multi-tier architecture, where smaller models are provided with a simplified interface that presents powerful tools in a more manageable way. Benchmarks show that the hybrid strategy used for tiny models results in an impressive 97% reduction in token usage while maintaining accuracy. By dynamically adjusting how models interact with tools based on their size, Tier promises to enhance the accessibility and efficiency of AI agents across various applications.