Arize AI released phoenix-evals version 2.6.0, enabling classification evaluators to accept AI SDK evaluation models, including TypeSafe AI's Jev1. When an evaluation model is passed, classification routes through experimental_evaluate as a single-choice question rather than generateObject. Results return a label and score but no explanation, as evaluation models do not generate text.
Arize Phoenix Evals Adds AI SDK Evaluation Model Support
Arize AI's phoenix-evals 2.6.0 lets classification evaluators accept AI SDK evaluation models like TypeSafe AI's Jev, routing via experimental_evaluate.