Adaptive inference with open-weight models achieves near-frontier performance at 1/3 cost

Tracer's Echo system demonstrates adaptive inference by combining open-weight models, achieving near-frontier performance at one-third the cost. Concrete benchmarks and an open-source eval observatory provide actionable insights for inference optimization and cost reduction.