1.1.0-beta.1 (2026-11-13)
Features Added
- Added multimodal embedding operations for text, queries, and document images, with dense, multi-vector, and combined results.
- Added inline and link delivery, output encoding and numeric precision selection, and embedding usage and warnings.
- Added flattened
GetEmbeddingsandGetEmbeddingsAsyncmethods; callers pass request settings directly rather than creating a request options object. - Added .NET overloads that accept anonymous objects or dictionaries as embedding inputs and return typed results, without requiring
EmbeddingInputinstances. Omitted input kinds default to text; PascalCase members and SDK content kinds are supported, and unknown or duplicate properties are rejected. - Added generic embedding result and usage models, with specialized multimodal implementations.
Breaking Changes
- The default service API version is now
2026-11-01-preview. SelectInferenceClientOptions.ServiceVersion.V2026_10_30explicitly to continue using the stable semantic reranking API.