A new developer endpoint called Echo, built by the team at Tracer, launched in public alpha on July 23, 2026, promising Claude-class output at roughly one-third the cost. Instead of betting on a single model, Echo pools open-weight models such as GLM-5.2 and Kimi K2.7 and decides, per request, how much compute to spend, which models should participate, and how to combine their outputs. It exposes all of this behind one OpenAI-compatible endpoint.

Try it: swap your base URL and route through a model pool

Echo is free during the alpha, with no payment method required. Sign up, create an API key from the dashboard, and point any OpenAI-compatible SDK at Echo's base URL to route chat, code, and agent calls through the pool instead of a single provider. Because the interface mirrors the OpenAI schema, most existing agent scaffolds and coding tools work with a one-line base-URL change. Start with a task you already benchmark, then compare Echo's combined answer against your current single-model setup using its published evaluation results.

Why It Matters

Model routing is becoming a real layer in the creator and builder stack. Rather than picking one frontier model and paying its rate for every call, an orchestration layer can send easy prompts to cheap models and reserve heavier combinations for hard ones. Echo claims it reached roughly the same aggregate quality as Claude Fable 5 at about a third of the cost, a pitch that landed it near the top of Hacker News the day it launched. If the numbers hold outside the vendor's own benchmarks, cost-sensitive builders get a cheaper default without rewriting their apps.

Key Details

Echo works from a pool of open-weight models, including Zhipu's GLM-5.2 and Moonshot's Kimi family, and combines their outputs per request rather than picking one. Its founder describes the design as recovering the advantage of an idealized system that knows in advance which models will help on a given problem, without actually having that information. A recurring finding: models that are weaker overall still contribute meaningfully as part of a combination, which is what makes pooling pay off. The alpha ships an OpenAI-compatible API, a conversational web interface for quick tests, and ten lifetime invites per account. There is no published GitHub repository yet, and the model list and routing internals are not fully documented, so treat the cost and quality claims as vendor benchmarks until independent tests appear. The open-weight foundation it leans on, models like Kimi, is exactly what makes an aggressive price point plausible.

What to Do Next

If you run agent or coding workloads at volume, spin up an Echo key and A/B a slice of real traffic against your current provider, watching both output quality and per-request cost. Keep your existing model as the fallback until you have your own numbers, since alpha endpoints change fast and availability is not guaranteed. Builders who care about routing economics should also track whether Echo publishes its model pool and opens the routing logic, which is what would turn a promising alpha into a dependable part of a production pipeline.