Tracer · 2026-07-23 · notable
Echo — Tracer's router blends open-weight LLMs at a third of Fable 5's cost
Echo is a public alpha from YC-backed lab Tracer that hides a pool of open-weight models (GLM-5.2, Kimi K2.7) behind one OpenAI-compatible endpoint and claims Claude Fable 5 quality at about 1/3 the inference cost.
One endpoint, many open-weight models, roughly a third of Claude Fable 5's inference cost on the tasks Tracer measured.
Pricing
| Quick — input | $3.00 / 1M tokens |
|---|---|
| Quick — output | $15.00 / 1M tokens |
| Focused — input | $6.00 / 1M tokens |
| Focused — output | $30.00 / 1M tokens |
| Deep — input | $10.00 / 1M tokens |
| Deep — output | $50.00 / 1M tokens |
| Alpha access · no card required during public alpha | $0 |
What is it?
Echo is the first product from Tracer, a YC-backed research lab focused on coordinated intelligence. It picks between a pool of open-weight models per request and returns the answer through a single OpenAI-compatible endpoint that developers can point Cursor, Cline, Continue, or their own SDK at.
How does it work?
Under the hood, Tracer routes each prompt across models such as GLM-5.2 and Kimi K2.7 and merges their answers instead of picking a single model up front. The API is stateless and exposes three tiers — Quick, Focused, and Deep — that Echo assigns automatically per request, with 907 stored evaluation rows across seven benchmark families backing the routing choices.
Why does it matter?
For teams that already lean on open-weight models, Echo turns model-picking into an API concern rather than a Cursor setting. Tracer claims Echo hits Claude Fable 5-class results at about a third of the cost — a real number to test if inference bills are the ceiling on how often agents can call an LLM.
Who is it for?
Developers running open-weight coding agents on a budget
Try it
export ECHO_API_KEY=... && curl https://echo.tracerml.ai/v1/chat/completions -d '{"model":"echo","messages":[...] }'