Sign in
← AI PulseSep 16, 2026

Wire · news · Multi-source brief

Vercel AI SDK Adds Experimental Evaluation Model Aliases and TypeSafe Provider

Vercel's AI SDK has introduced experimental evaluation model aliases and registry resolution, alongside a new TypeSafe provider for experimental_evaluate that supports various question types and structured rubrics.

By Illumora Editorial

Source · Sep 16, 2026, 9:31 PM · On Illumora · Sep 16, 2026, 9:43 PM

Media from the primary source — shown here so you can stay on Illumora.

Synthesized from multiple allowlisted primaries on the same event. Lanes →

Brief drafted by Illumora’s editorial model from the linked primary source. Ops desk reviews flagged pieces. How we write →

Read the source →Vercel AI SDK Changelog — Release @ai-sdk/workflow-harness@1.0.114 · vercel/ai
Save

The Vercel AI SDK has been updated to include experimental evaluation model aliases and registry resolution. This update allows customProvider to accept evaluationModels and exposes evaluationModel through registries. The experimental_evaluate function can now accept string IDs when a default evaluation-capable provider is explicitly configured.

Separately, the @ai-sdk/typesafe-ai@3.0.0 package introduces a TypeSafe provider for experimental_evaluate. This provider is designed to support native Choice, Score, and Boolean questions within a single request, along with structured JSON rubrics, probability distributions, confidence metadata, and workflow serialization.

Key Points

  • The Vercel AI SDK now supports experimental evaluation model aliases and registry resolution.
  • customProvider can accept evaluationModels.
  • Registries expose evaluationModel.
  • experimental_evaluate accepts string IDs when an evaluation-capable default provider is configured.
  • Evaluation does not implicitly fall back to Gateway.
  • The @ai-sdk/typesafe-ai@3.0.0 package adds a TypeSafe provider for experimental_evaluate.
  • This TypeSafe provider supports native Choice, Score, and Boolean questions in one request.
  • It also supports structured JSON rubrics, probability distributions, confidence metadata, and workflow serialization.

Context

According to Vercel, model-resolution errors now specifically identify evaluationModel, while stable provider contracts remain unchanged. The @ai-sdk/typesafe-ai@3.0.0 update, released on September 16, also includes updates to dependencies such as @ai-sdk/provider-utils@5.0.42 and @ai-sdk/provider@4.0.16.

Why It Matters

These updates provide developers with more explicit control and structured capabilities for evaluating AI models within the Vercel AI SDK, potentially improving the reliability and interpretability of evaluation results.

What To Do

  • Review the Vercel AI SDK documentation for details on experimental_evaluate and evaluationModels.
  • Test the new TypeSafe provider for experimental_evaluate if your application requires structured evaluation questions or rubrics.
  • Note that evaluation will not implicitly fall back to Gateway, requiring explicit configuration.
  • Compare the new evaluation model aliases and registry resolution features with existing evaluation workflows.