Sign in
← AI PulseSep 16, 2026

Wire · news · Multi-source brief

Vercel AI SDK Introduces Experimental Evaluation Model Aliases and Tool Search

The Vercel AI SDK has added experimental evaluation model aliases and registry resolution, alongside new tool search capabilities and a TypeSafe provider for experimental_evaluate.

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/vue@4.0.104 · vercel/ai
Save

The Vercel AI SDK has introduced several updates, including experimental evaluation model aliases and registry resolution, as detailed in recent changelog entries for @ai-sdk/vue@4.0.104 and @ai-sdk/workflow@2.0.35. These changes enable customProvider to accept evaluationModels and registries to expose evaluationModel.

Additionally, the SDK now supports tool search with direct tool calling and a native tool search tool. A new TypeSafe provider for experimental_evaluate was also added, supporting various question types and structured JSON rubrics, as noted in the @ai-sdk/typesafe-ai@3.0.0 release on September 16.

Key Points

  • The Vercel AI SDK now includes experimental evaluation model aliases and registry resolution.
  • customProvider accepts evaluationModels and registries expose evaluationModel.
  • The experimental_evaluate function accepts string IDs when an evaluation-capable default provider is explicitly configured.
  • Evaluation does not implicitly fall back to Gateway, and model-resolution errors identify evaluationModel.
  • The SDK supports tool search with direct tool calling and a native tool search tool.
  • A TypeSafe provider for experimental_evaluate supports native Choice, Score, and Boolean questions in one request.
  • The TypeSafe provider also supports structured JSON rubrics, probability distributions, confidence metadata, and workflow serialization.

Context

According to the Vercel AI SDK changelog, these updates aim to enhance the flexibility and capabilities of the SDK for developers. The introduction of experimental evaluation models and tool search features provides more granular control and advanced functionalities for AI applications. The @ai-sdk/typesafe-ai@3.0.0 release, specifically, focuses on robust evaluation mechanisms, as stated by Vercel.

Why It Matters

These updates offer builders more sophisticated tools for integrating and evaluating AI models within their applications. The ability to define and resolve evaluation models, coupled with enhanced tool search, can streamline development workflows and improve the reliability of AI-powered features. The TypeSafe provider's structured evaluation capabilities could lead to more precise and consistent model assessments.

What To Do

  • Review the Vercel AI SDK changelog for @ai-sdk/vue@4.0.104, @ai-sdk/workflow@2.0.35, and @ai-sdk/typesafe-ai@3.0.0.
  • Explore the use of customProvider with evaluationModels for custom evaluation setups.
  • Test the new tool search capabilities with direct tool calling in your projects.
  • Investigate the TypeSafe provider for experimental_evaluate to implement structured evaluations.