Benchmarks & research
Cribl's first open decision model is a LoRA adapter for Qwen3.5-4B that gives a probability for every option you supply, meant as a base for telemetry, IT and security models.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
Only the adapter is published, so the base model is downloaded separately at the exact revision Cribl names. The card says exact outputs depend on Cribl's candidate-scoring readout and local server, which are not in the repository. It is a separate work from Cribl's earlier post testing Jev on telemetry.
Cribl-decision-1.0 suits sorting large numbers of log events, support tickets or requests into answers you list yourself, on your own machines. You give it the item, a question and the allowed answers, and it says how likely each one is. Cribl advises checking those numbers on your own data and keeping a person in charge of high-stakes calls.