Tracer AI opens public alpha of Echo, an adaptive router over open-weight models

Tracer AI, a Y Combinator-backed research lab, put its Echo adaptive router into public alpha on July 23, claiming it matches the quality of a frontier model at roughly one-third the inference cost.

Echo is an OpenAI-API-compatible endpoint that coordinates multiple open-weight models on a per-request basis for chat, code and agent tasks. Instead of running every prompt through one large model, it routes each request to whichever open-weight model it scores as best suited for the task — a bid to cut the cost of running AI applications without giving up much quality.

The router draws on open-weight models including GLM-5.2 and Kimi K2.7. Tracer says Echo delivers results comparable to Anthropic’s Claude Fable across its published task mix at about a third of the total cost, and that Echo outperformed every individual open-weight model it tested.

Those numbers come entirely from Tracer’s own evaluation, which it calls the Echo Eval Observatory, and have not been independently benchmarked. The lab itself hedges the pitch, calling the results “scoped evidence, not a claim that Echo wins every task.”

Model routing is a crowded idea, and cost-versus-quality claims are notoriously sensitive to which tasks and price assumptions a vendor picks. The public alpha lets outside developers run their own prompts, which is the only way the one-third-cost claim gets tested against workloads Tracer did not choose.