Echo Claims Fable Parity at a Third the Cost with Open Models

Most model launches ask you to buy into a new weights file. Echo, which surfaced on Hacker News as a Show HN from a builder posting as adam_rida, is pitching something different: one AI system stitched together from a pool of open-weight models, deciding per request how much compute to spend, which models to call, and how to combine their answers. The headline claim is Fable-level results at roughly one-third the inference cost.

The mechanic, as the post describes it, is that some prompts only need a small model, others benefit from several models working on different parts, and Echo makes that call on the fly. The builder writes that Echo "consistently performed better than the best individual model in its pool" while reaching roughly the same aggregate result as Fable. Read that carefully: matching a frontier tier by ensembling cheaper models is a different result from any single model in the pool being frontier-class.

The transparency move is a public evaluation dashboard hosting 907 examples across seven benchmark families, with prompts, outputs, grades, and cost records visible without a login. That is more raw material than most launches offer, and it is the right place to point sceptical readers. The honest caveat, which commenters raised almost immediately, is that the routing policy itself is closed: as the post puts it, "Echo does not disclose its per-request routing decision because that policy is the product," so an enterprise buyer cannot tell which model handled which request. Others flagged sign-up friction, calling a message box that redirects to signup "a classic dark pattern," and questioned whether some of the benchmarks used are saturated enough to hide real differences. The creator also says they are still testing coding and agentic tasks, which is where developer buyers care most.

What the reporting does not give you is the model roster, the exact benchmark families used, or any independent replication of the cost math. Take the parity claim as a first-party assertion with a dashboard attached, not settled fact.

If the numbers survive outside probing, the group this most benefits is teams already burning frontier-API budget who would happily route easier requests to something cheaper. That is the wedge to watch.