Show HN: Echo – Fable-level results at 1/3 the cost using open-weight models

I’ve been building Echo: one adaptive model built entirely from a pool of open-weight models, including GLM-5.2, Kimi K2.7, and many others.

It started with a surprising result: a perfect oracle over the pool beats the state of the art on every benchmark we tested.

Echo is my attempt to make that practical.

On our first internal task mix, it consistently beat the best individual model in the pool and reached Fable-level results at roughly 1/3 of the total inference cost.

It’s still early. Echo already matches the state of the art on some coding benchmarks, while closing the remaining gaps is one of our main focuses now.

You can try Echo in the chat, or plug its OpenAI-compatible API into OpenCode and other compatible tools.

New accounts include $10 of free inference credits.

Try it on something hard!

Evaluation methodology, results and limitations: echo.tracerml.ai/eval