The model selects among supplied options and returns a probability distribution. It stands out on tool selection and intent classification, but is weaker on open-ended reasoning and some other tests.

The Apache-2.0-licensed model uses BF16 weights. Application-specific calibration is needed before its scores influence consequential decisions.

لماذا يهم

It offers an interface for fixed-choice routing and classification systems, though the reported benchmark score is not independent verification.

أدلة

شيء لتجربته

For a hypothetical prototype, plan how to evaluate its probability distribution against an existing classifier on a limited label set.