ARC 1.0: An Open-Source Decision Model for Routing AI Agents
ARC 1.0 has been released as an open-source, 0.6B adaptive routing controller under the Apache 2.0 license to help AI applications choose optimal models and reasoning efforts. It analyzes conversation history to generate routing signals that applications combine with pricing and policy constraints. This tool addresses a major cost and performance optimization bottleneck in local LLM deployments by dynamically balancing workloads between expensive cloud models and efficient local alternatives. It provides developers with fine-grained control over multi-agent architecture execution costs. ARC 1.0 is a compact 0.6B parameter model distributed under the permissive Apache 2.0 open-source license. It operates by reading an ongoing agent conversation and outputting specific signals to guide subsequent model and reasoning effort selections.
## BACKGROUND
Model routing in AI refers to the practice of dynamically directing different queries to different LLM instances based on task complexity, helping balance operational costs with performance needs. Local LLMs run directly on user hardware, offering privacy and zero per-token API costs, while cloud models provide superior reasoning capabilities at a higher financial expense.