Sign in
← AI PulseSep 16, 2026

Wire · news · Multi-source brief

Vercel AI SDK Adds TypeSafe Provider for Evaluation, Enhances Tool Discovery

The Vercel AI SDK has introduced a TypeSafe provider for experimental_evaluate, enabling native support for Choice, Score, and Boolean questions, alongside structured JSON rubrics and workflow serialization.

By Illumora Editorial

Source · Sep 16, 2026, 6:26 PM · On Illumora · Sep 16, 2026, 6:39 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/luma@3.0.43 · vercel/ai
Save

The Vercel AI SDK has released updates across several packages, including @ai-sdk/luma@3.0.43, @ai-sdk/huggingface@2.0.50, @ai-sdk/hume@3.0.42, and @ai-sdk/klingai@4.0.44. A significant update is the addition of a TypeSafe provider for experimental_evaluate within the @ai-sdk/typesafe-ai@3.0.0 package.

Key Points

  • The TypeSafe provider for experimental_evaluate supports native Choice, Score, and Boolean questions in a single request.
  • It also includes structured JSON rubrics, probability distributions, and confidence metadata.
  • Workflow serialization is a new feature supported by the TypeSafe provider.
  • The AI SDK now features mid-conversation tool discovery and updates for code-mode.
  • The rawInput in output-error UI message parts has been deprecated.
  • Default Node.js downloads are protected by DNS validation and connection pinning.

Context

According to the Vercel AI SDK changelog, these updates aim to enhance the evaluation capabilities and overall robustness of the SDK. The TypeSafe provider specifically addresses structured evaluation needs, while other changes improve tool integration and security measures within the SDK.

Why It Matters

These updates provide developers with more structured and secure ways to evaluate AI model outputs and integrate tools within their applications. The experimental_evaluate feature, with its support for various question types and JSON rubrics, offers a more rigorous approach to assessing model performance and reliability.

What To Do

  • Review the documentation for the experimental_evaluate function and the new TypeSafe provider.
  • Explore the new capabilities for mid-conversation tool discovery in code-mode.
  • Note the deprecation of rawInput in output-error UI message parts and adjust existing implementations if necessary.
  • Verify that Node.js downloads within your environment benefit from the enhanced DNS validation and connection pinning.