Clef: Open-weight decision models, and new RL fine-tuning platform
Thread
Unofficial Hacker News client; not affiliated with Y Combinator.
Clef: Open-weight decision models, and new RL fine-tuning platform
Unofficial Hacker News client; not affiliated with Y Combinator.
croemer · · focus · HN ↗
handfuloflight · · focus · HN ↗
croemer · · focus · HN ↗
mpolichette · · focus · HN ↗
I'd love an privacy first on-device model i could use in iOS.
okpatil · · focus · HN ↗
At0M: A 60M local Jev at 16 ms latency and 79% accuracy on Typed Decision
<a href="https://at0m.pienomial.com/" rel="nofollow">https://at0m.pienomial.com/ <a href="https://news.ycombinator.com/item?id=49920350">https://news.ycombinator.com/item?id=49920350
itsmeduncan · · focus · HN ↗
[dead]
okpatil · · focus · HN ↗
At0M: A 60M local Jev at 16 ms latency and 79% accuracy on Typed Decision
<a href="https://at0m.pienomial.com/" rel="nofollow">https://at0m.pienomial.com/ <a href="https://news.ycombinator.com/item?id=49920350">https://news.ycombinator.com/item?id=49920350
networked · · focus · HN ↗
You can list the decision models available on OpenRouter: <a href="https://openrouter.ai/models?output_modalities=decisions" rel="nofollow">https://openrouter.ai/models?output_modalities=decisions. There are currently nine.
My AI harness uses Jev or a compatible model to approve shell commands that look safe without asking the user. (This is one security layer.) A quick benchmark (<a href="https://github.com/dbohdan/strument/tree/c2320ca142f77495ecdcadd3cd7a4c9b9445369f/doc/experiments/2026-10-decision-models" rel="nofollow">https://github.com/dbohdan/strument/tree/c2320ca142f77495ecd...) has shown that of five models on OpenRouter only Liquid D1 performs similar to Jev for this.