The Vercel AI SDK has integrated a new Z.AI provider, enabling access to GLM chat completions, streaming capabilities, reasoning, tool use, and multimodal inputs. This update was part of the @ai-sdk/zai@2.0.0 release, which occurred on August 26.
Key Points
- The Z.AI provider supports GLM chat completions.
- The integration includes streaming, reasoning, and tool use functionalities.
- Multimodal inputs are supported through the Z.AI provider.
- The GLM-5.3-Flash model is specifically supported within the Z.AI provider and AI Gateway.
- The @ai-sdk/moonshotai package received updates for Moonshot V1 auto and vision-preview model IDs.
- Moonshot chat log probabilities and partial mode for final assistant message prefixes are now supported.
- The @ai-sdk/amazon-bedrock package forwards the Anthropic option for disabling parallel tool use.
Context
According to the Vercel AI SDK changelog, the Z.AI provider was added as a major change in the @ai-sdk/zai@2.0.0 release. This release also included a backport to add GLM-5.3-Flash model support to the Z.AI provider and AI Gateway. Separately, the @ai-sdk/moonshotai package was updated to include first-class support for Moonshot V1 auto and vision-preview model IDs, while preserving custom and retired model ID support. It also gained support for partial mode for final assistant message prefixes and chat log probabilities.
Why It Matters
This expansion of the Vercel AI SDK provides developers with more options for integrating advanced AI capabilities, including multimodal interactions and specific GLM models, into their applications. The updates to Moonshot and Amazon Bedrock integrations also enhance flexibility and control over model behavior and output.
What To Do
- Review the @ai-sdk/zai@2.0.0 release notes for details on Z.AI provider integration.
- Test the GLM-5.3-Flash model through the Z.AI provider and AI Gateway.
- Explore the new Moonshot V1 auto and vision-preview model IDs and features in @ai-sdk/moonshotai.
- Note the updated handling of Anthropic parallel tool use options in @ai-sdk/amazon-bedrock.
