Vercel ai@7.0.174 Adds Batch Output Support and Extraction Model Improvements
Vercel's ai SDK version 7.0.174 introduces output support for batch operations, improves extraction model availability, and fixes memory usage issues during metadata validation.
What changed?
Vercel released ai@7.0.174 with several enhancements and fixes: - The startBatch function now supports an output argument, and getBatchResults can parse this output. - Extraction models can now be used in provider registries, custom providers, and specified by string model IDs. - The streamText method now provides an empty tool set to transforms if no tools are given, preventing undefined tool issues. - Validation of deeply nested provider or tool metadata has been optimized to reduce excessive memory and CPU usage.
Why does it matter to an everyday developer?
These changes improve flexibility for developers using the batch API by allowing direct control and inspection of batch outputs. Extraction model support expands the types of models that can be registered and integrated, enabling new types of AI-powered data extraction workflows. The fix to tool set handling in streamText reduces the likelihood of runtime errors in custom transform scenarios. The metadata validation change addresses a performance bottleneck that could cause worker crashes in cases of deeply nested data, improving reliability of AI processes in production environments.
What can the developer do now?
Developers using batch processing can now utilize the output argument in startBatch and benefit from improved result parsing with getBatchResults. They can register and integrate extraction models more easily in provider registries or custom providers. Those building custom transforms with streamText will observe safer default behavior around tools. Teams with complex metadata structures will see lower memory usage and greater stability in validation routines. Update to ai@7.0.174 to start using these improvements.
