@arizeai/phoenix-evals@2.6.0
Arize Phoenix evals v2.6.0 introduces support for AI SDK evaluation models like TypeSafe's Jev, routing classification via experimental evaluate without explanations.
Why this signal matters
Arize Phoenix released version 2.6.0 of @arizeai/phoenix-evals . The release updates classification evaluators to accept AI SDK evaluation models, citing TypeSafe's Jev as an example. When configured with an evaluation model via the model parameter, execution switches from generateObject to an experimental evaluate flow formatted as a single choice question. Output results include a label and score , but exclude an explanation because evaluation models do not generate text.
Actionable summary
In @arizeai/phoenix-evals@2.6.0, classification evaluators support AI SDK evaluation models (e.g., TypeSafe's Jev). Evaluators route through experimental evaluate as single-choice questions rather than generateObject, yielding label and score but omitting explanation fields.
- Agent usefulness
- 77/100
- Confidence
- 96%
- Canonical data
- JSON + Markdown
What builders should check
- Verify downstream evaluation workflows do not depend on the 'explanation' field when passing AI SDK evaluation models to classification evaluators.
- Update to @arizeai/phoenix-evals@2.6.0 to leverage AI SDK evaluation models such as TypeSafe's Jev.