Vercel's AI SDK has received updates across several packages, including @ai-sdk/svelte@5.0.85, @ai-sdk/togetherai@3.0.42, and @ai-sdk/tui@1.0.86. These releases, which update the core ai package to 7.0.85, address image generation cost calculations and enhance provider capabilities, particularly for Anthropic models.
Key Points
- The AI SDK now correctly sums Gateway image-generation costs across split requests.
- The Anthropic provider preserves native message batch request counts in provider metadata.
- The Anthropic provider now supports the full language-model option surface in batch requests.
- Individual image generation calls are now exposed in the AI SDK.
- Amazon Bedrock now enables Anthropic reasoning budgets for application inference profile ARNs.
- The AI SDK exposes parsed structured output in
streamTextend callbacks. - Amazon Bedrock accepts citation deltas in streaming responses.
Context
According to the Vercel AI SDK Changelog, these updates include patch changes and dependency updates across various SDK components. The changes aim to refine cost reporting for image generation and improve the integration with specific providers like Anthropic and Amazon Bedrock. Previous updates, as noted in the @ai-sdk/svelte@5.0.84 release, also introduced features such as exposing parsed structured output in streamText end callbacks and accepting citation deltas in Amazon Bedrock streaming responses.
Why It Matters
These updates enhance the accuracy of cost tracking for image generation and expand the functional parity of batch requests with Anthropic's native API. Builders can leverage these changes for more precise resource management and broader access to language model options within the SDK.
What To Do
- Note the updated cost summation for Gateway image generation if using that feature.
- Review the Anthropic provider documentation for the expanded language-model option surface in batch requests.
- Test the exposure of individual image generation calls in your applications.
- Check Amazon Bedrock configurations for enabling Anthropic reasoning budgets with application inference profile ARNs.
