•
Discovery API:aiplatform.v1beta1↗
•
Caught 7 days ahead of release
Released (7d lead)Breaking Change
Docs

AI Platform v1beta1: Major Breaking Changes & New File/Embedding APIs

Modified API Methods (15):

This Vertex AI v1beta1 update introduces several critical breaking changes. The entire evaluationTasks.operations resource and its methods (delete, get, list, wait) have been removed, requiring immediate code updates for any integrations relying on these. Furthermore, the continuationToken property in GenaiVertexV1beta1Interaction, which was recently added to support resuming long-running content generation, has been removed, rolling back this feature. Developers will need to revert any code changes made to utilize this token.

Additional breaking changes include several parameters becoming strictly required: name for projects.locations.publishers.files.delete, download, and get; parent and publishersId for projects.locations.publishers.files.list; and model for projects.locations.publishers.models.batchEmbedContents. Also, the aiApplication property in GoogleCloudAiplatformV1beta1Agent and multiple fields within GoogleCloudAiplatformV1beta1ClaudeFile (including created_at, downloadable, expires_at, filename, id, mime_type, size_bytes, and type) are now read-only.

On the feature front, a new files resource has been added under projects.locations.publishers, providing delete, download, get, and list methods for managing Claude files. The list method supports paginated or batch fetching by ids. The projects.locations.publishers.models resource now includes a batchEmbedContents method, enabling efficient batch embedding of content. The projects.locations.agents.create method gains an optional idpNonce parameter for enhanced session and device non-repudiation. Video generation experiments (CloudAiLargeModelsVisionGenerateVideoExperiments) are extended with a new perfEstimationConfig for performance estimation, including BoundingBox and CharacterReference details. Finally, GoogleCloudAiplatformV1beta1GenerateContentRequest now supports an outputConfig field, allowing you to specify an outputUriPrefix for generated content.