# What changed in @ai-sdk/azure > 26 changes in @ai-sdk/azure 3 to 4 from its official release notes, upgrade guides and advisories: 3 breaking, 2 deprecations, 16 new APIs and options. Each one cited. Sources checked 2026-10-17. HTML: https://unstale.sh/npm/@ai-sdk/azure/ To see which of these apply to a repository, run `npx @unstale/cli init` in it. ## @ai-sdk/azure 4 ### Breaking changes - Azure Speech voice-reset errors from unknown voices or unsupported styles now report status 400 instead of 502, preventing retries for client input errors. (3.0.134, 4.0.91) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%404.0.91), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%403.0.134) - Unknown MAI voices or styles now produce descriptive, non-retryable errors. (3.0.133, 4.0.90) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%404.0.90), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%403.0.133) - Rejects resourceName values that are not single DNS labels, preventing malformed values from rewriting the request host. (3.0.131, 4.0.84) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%403.0.131), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%404.0.84) ### Deprecations - Deprecated in favor of azure.tools.webSearch(). (4.0.104) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%404.0.104) - Deprecated or ineffective DeepSeek sampling options now produce warnings and are omitted, while penalty sampling remains supported for Azure-hosted DeepSeek models. (3.0.107, 4.0.49) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%404.0.49), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%403.0.107) ### New APIs and options - Adds support for MAI-Image models through the MAI image API. (3.0.135, 4.0.92) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%404.0.92), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%403.0.135) - Adds streaming transcription support through the MAI realtime API. (4.0.90) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%404.0.90) - Adds speech generation support for MAI-Voice-2.1 through the MAI realtime API. (4.0.90) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%404.0.90) - Adds speech generation support for MAI-Voice-2.1-Flash through the MAI realtime API. (4.0.90) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%404.0.90) - Adds the maiBaseURL setting for MAI realtime API integration. (4.0.90) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%404.0.90) - Adds providerOptions.azure.api: 'mai' to select the MAI realtime API. (4.0.90) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%404.0.90) - Adds the webSocket setting for MAI realtime API integration. (4.0.90) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%404.0.90) - Adds MAI-Transcribe-2 file transcription through the Azure Speech API with diarization, timestamps, transcript styles, locale forcing, and phrase lists. (3.0.128, 4.0.83) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%404.0.83), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%403.0.128) - Adds azure.transcriptionModel as an alias of azure.transcription. (3.0.128, 4.0.83) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%404.0.83), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%403.0.128) - Adds an Azure API provider option that overrides model-based routing for speech generation. (3.0.131, 4.0.83) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%403.0.131), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%404.0.83) - Adds a setting for configuring the Azure Speech endpoint. (3.0.128, 4.0.83) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%404.0.83), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%403.0.128) - Exports OpenAIResponsesSystemMessageOptions for typed message-level reasoning effort updates in Azure Responses models. (2.0.137, 3.0.126, 4.0.79) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%404.0.79), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%403.0.126), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%402.0.137) ### Behavior and default changes - Azure AI Foundry Responses requests now include explicit message item types. (2.0.132, 3.0.118, 4.0.66) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%404.0.66), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%403.0.118), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%402.0.132) ### Fixes and other changes - Hostname parts are now validated using isValidHostnamePart, affecting hostname validation for configured Azure endpoints. (4.0.87) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%404.0.87) - Uses a standards-compliant User-Agent header. (4.0.84) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%404.0.84) - Constructs OpenAI v1 URLs for Azure AI Foundry and Cognitive Services hostnames while preserving complete v1 and Foundry project base URLs. (2.0.129, 3.0.113, 4.0.55) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%404.0.55), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%403.0.113), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%402.0.129) ## @ai-sdk/azure 3 ### New APIs and options - Adds support for MAI-Voice-2.1 and MAI-Voice-2.1-Flash speech generation. (3.0.133) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%403.0.133) - Adds MAI-Voice-2-Flash and MAI-Voice-2 speech generation through Azure Speech text to speech with voice, output format, speed, and style options. (3.0.131) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%403.0.131) - Adds azure.speechModel() as an alias of azure.speech(). (3.0.131) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%403.0.131) - Adds an Azure provider option to override model-based API routing for transcription. (3.0.128) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%403.0.128) ### Behavior and default changes - Requests route to the Azure Speech API by default and no longer send rejected segment timestamp defaults. (3.0.131) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/azure%403.0.131)