Azure SDK for Python (March 2026)
The Azure SDK team is pleased to announce our March 2026 client library releases.
36 packages released this month.
Stable Packages (13)
-
AI Evaluation
-
AI Projects
-
App Configuration Provider
-
Azure Content Understanding in Foundry Tools
-
Cosmos DB
-
Resource Management - Elastic SAN
-
Resource Management - Artifactsigning
-
Resource Management - Batch
-
Resource Management - Data Factory
-
Resource Management - Disconnectedoperations
-
Resource Management - NetApp Files
-
Resource Management - Service Fabric Managed Clusters
-
Text Analytics
Patch Updates (5)
-
AI Evaluation
-
AI Projects
-
Azure Content Understanding in Foundry Tools
-
Core - Client - Core
-
Identity
Beta Packages (12)
-
AI Agent Server - Core
-
AI Projects
-
Confidential Ledger
-
Mixed Reality Authentication
-
Resource Management - Commerce
-
Resource Management - Communication
-
Resource Management - Computebulkactions
-
Resource Management - Connected Cache
-
Resource Management - Managedops
-
Resource Management - Mixed Reality
-
Resource Management - Resource-Policy
-
unknown
Release highlights
AI Agent Server - Core 1.0.0b15 Changelog
Other Changes
- Upgraded starlette>=1.0.0rc1.
AI Agent Server - Core 1.0.0b16 Changelog
Other Changes
- Upgraded to support agent-framework v1.0.0rc3.
AI Evaluation 1.15.1 Changelog
Bugs Fixed
- Red Team Agent Scenario Integration: integrated PyRIT’s FoundryScenario for attack orchestration with Azure-specific scoring and result processing.
- Fixed total tokens calculation errors in evaluation results.
- Fixed red team SDK run status updates not firing consistently, preventing runs from being stuck as “Running” in the UI.
AI Evaluation 1.15.2 Changelog
Bugs Fixed
- Fixed batch evaluation to report per-line errors only on the rows that failed.
AI Evaluation 1.15.3 Changelog
Bugs Fixed
- Fixed Foundry red team double-evaluation that caused ~3x scan latency by removing redundant evaluation_processor.evaluate() call.
- Fixed lost red team attack details where baseline overwrites wiped evaluation results.
- Fixed incorrect per_testing_criteria that showed attack strategies alongside risk categories.
- Fixed metadata leak of internal scorer fields (attack_success, attack_strategy, score) into results.json.
- Improved error logging with run_id, display_name, and full stack traces for red team scan failures.
AI Evaluation 1.16.0 Changelog
Bugs Fixed
- Fixed
UnicodeDecodeErroron Windows when reading red team JSONL files containing non-ASCII characters (UnicodeConfusable strategy, CJK languages) by adding explicitencoding="utf-8"to all file open calls in the result processing path. - Fixed
NotFoundError: 404when usingmodel_configdict target with Foundry-style endpoints (*.services.ai.azure.com) by appending/openai/v1to the endpoint URL for PyRIT compatibility. - Fixed red team scan status stuck at
in_progressin results.json despite the scan completing, by treating leftoverpendingentries asfailed. - Fixed
ungrounded_attributesrisk category being silently skipped due to a cache key mismatch (isavsungrounded_attributes) in the Foundry execution path. - Fixed RAI evaluation service errors (
ServiceInvocationException) incorrectly inflating attack success rate by treating error responses as undetermined instead of attack success. - Fixed Foundry red team double-evaluation that caused ~3x scan latency by removing redundant evaluation_processor.evaluate() call.
- Fixed lost red team attack details where baseline overwrites wiped evaluation results.
- Fixed incorrect per_testing_criteria that showed attack strategies alongside risk categories.
- Fixed metadata leak of internal scorer fields (attack_success, attack_strategy, score) into results.json.
- Improved error logging with run_id, display_name, and full stack traces for red team scan failures.
- Fixed httpx read timeout errors during red team scans by configuring explicit HTTP timeout on PyRIT OpenAIChatTarget instances (default: 180s, configurable via
_http_timeoutscan kwarg). - Fixed
InvalidInputErrorcrash when evaluating agent tool output containing unresolvable relative image paths from Document Intelligence indexed PDFs, by gracefully falling back to text representation. - Fixed
InvalidInputErrorcrash on markdown image references with empty alt text (e.g.) from Document Intelligence by allowing empty alt text in the image URL parser.
AI Projects 2.0.0b4 Changelog
Features Added
- Tracing: included agent ID in response generation traces when available.
- Tracing: Added support for opt-in trace context propagation.
Breaking changes
- A Responses call on OpenAPI client (
openai_client.responses.create()) that uses an Agent reference, now needs to specifyextra_body={"agent_reference": {"name": agent_name, "type": "agent_reference"}}instead ofextra_body={"agent": {"name": agent_name, "type": "agent_reference"}}. - Agent methods
.agents.create(),.agents.create_from_manifest(),.agents.update()and.agents.update_from_manifest()were removed. Use the remaining methods.agents.create_version()and.agents.create_version_from_manifest()instead. - To align with OpenAI naming conventions, use “Tool” suffix for class names describing Azure tools that are generally available (stable release):
- Rename class
AzureAISearchAgentTooltoAzureAISearchTool. - Rename class
AzureFunctionAgentTooltoAzureFunctionTool. - Rename class
BingGroundingAgentTooltoBingGroundingTool. - Rename class
OpenApiAgentTooltoOpenApiTool. - To align with OpenAI naming conventions, use “PreviewTool” suffix for class names describing Azure tools in preview:
- Rename class
A2ATooltoA2APreviewTool. - Rename class
BingCustomSearchAgentTooltoBingCustomSearchPreviewTool. - Rename class
BrowserAutomationAgentTooltoBrowserAutomationPreviewTool. - Rename class
MemorySearchTooltoMemorySearchPreviewTool. - Rename class
MicrosoftFabricAgentTooltoMicrosoftFabricPreviewTool. - Rename class
SharepointAgentTooltoSharepointPreviewTool. - Other class renames:
- Rename class
PromptAgentDefinitionTexttoPromptAgentDefinitionTextOptions - Rename class
EvaluationComparisonRequesttoInsightRequest - To use Workflow Agents, which are still in preview, you now need to set an additional input
argument
foundry_features=FoundryFeaturesOptInKeys.WORKFLOW_AGENTS_V1_PREVIEWwhen calling.agents.create_version(). - To use Hosted Agents, which are still in preview, you now need to set an additional input
argument
foundry_features=FoundryFeaturesOptInKeys.HOSTED_AGENTS_V1_PREVIEWwhen calling.agents.create_version(). - To use
.evaluation_rules.create_or_update()withHumanEvaluationPreviewRuleAction, you now need to set an additional input argumentfoundry_features=FoundryFeaturesOptInKeys.EVALUATIONS_V1_PREVIEW. - Operation sets that are still in preview now have the “.beta” subclient in their call path. So for example
project_client.memory_stores.create()has changed toproject_client.beta.memory_stores.create(). Similarly for the operation sets:evaluators,insights,evaluation_taxonomies,schedulesandred_teams. - The method
begin_update_memories()in Memory Stores operation now accept optionalitemsof typeList[dict[str, Any]]instead ofList[ItemParam]. Similarly foritemsin methodsearch_memories(). As a result around 100 classes that are derived fromItemParamwere removed as they are no longer used by the client library. - Tracing instrumentation, is an experimental preview feature, now requires explicitly opt in by setting the environment variable:
AZURE_EXPERIMENTAL_ENABLE_GENAI_TRACING=true - Tracing: workflow actions in conversation item listings are now emitted as “gen_ai.conversation.item” events (with role=”workflow”) instead of “gen_ai.workflow.action” events in the list_conversation_items span.
- Tracing: response generation span names changed from “responses {model_name}” to “chat {model_name}” for model calls and from “responses {agent_name}” to “invoke_agent {agent_name}” for agent calls.
- Tracing: response generation operation names changed from “responses” to “chat” for model calls and from “responses” to “invoke_agent” for agent calls.
- Tracing: response generation uses gen_ai.input.messages and gen_ai.output.messages attributes directly under the span instead of events.
- Tracing: agent creation uses gen_ai.system_instructions attribute directly under the span instead of an event. Note that the attribute name is gen_ai.system_instructions not gen_ai.system.instructions.
- Tracing: “gen_ai.provider.name” attribute value changed to “microsoft.foundry”.
- Tracing: the format of the function tool call related traces in input and output messages changed to {“type”: “tool_call”, “id”: “…”, “name”: “…”, “arguments”: {…}} and {“type”: “tool_call_response”, “id”: “…”, “result”: “…”}
AI Projects 2.0.0 Changelog
Breaking Changes
- Input argument
foundry_featureswas removed from all methods that supported it. Use the newallow_previewinstead on client constructor (see above). - Class
TextResponseFormatConfigurationrenamed toTextResponseFormat. - Class
TextResponseFormatConfigurationResponseFormatTextrenamed toTextResponseFormatTest. - Class
TextResponseFormatConfigurationResponseFormatJsonObjectrenamed toTextResponseFormatJsonObject. - Class
CodeInterpreterContainerAutowas renamed toAutoCodeInterpreterToolParam, and has a new optional propertynetwork_policyof typeContainerNetworkPolicyParam. - class
ImageGenActionEnumwas renamed toImageGenAction. - Rename
ToolChoiceParamType.WEB_SEARCH_PREVIEW2025_03_11toToolChoiceParamType.WEB_SEARCH_PREVIEW_2025_03_11. - Rename
RankerVersionType.DEFAULT2024_11_15toRankerVersionType.DEFAULT_2024_11_15. - Rename method
.beta.evaluators.list_latest_versions()to.beta.evaluators.list(). - Rename property
idon classInsighttoinsight_id. - Rename property
idon classScheduletoschedule_id. - Rename input argument
idtoinsight_idin.beta.insights.get()method. - Rename input argument
idtoschedule_idin.beta.schedulesmethods. - Updated datetime-typed fields (
start_time,end_time,trigger_at,trigger_time,created_at,modified_at) acrossCronTrigger,RecurrenceTrigger,OneTimeTrigger,ScheduleRun, andEvaluatorVersionclasses fromstrtodatetime.datetimewith format=”rfc3339”.
Features Added
- To enable preview (beta) operations, a new optional boolean input argument named
allow_previewwas added to the constructor ofAIProjectClient. Caller must set it to True to opt-in to preview features. This includes creating an Hosted Agent or Workflow Agent. Methods on the.betasub-client (for example.beta.memory_stores.create()) do not require settingallow_preview=Truesince it’s implied by the sub-client name. When preview features are enabled, the client libraries sends the HTTP request headerFoundry-Featureswith the appropriate value in all relevant calls to the service.
AI Projects 2.0.1 Changelog
Bugs Fixed
- Fix custom Memory Stores LRO poller operation to add the missing
required
"Foundry-Features": "MemoryStores=V1Preview"HTTP request header.
App Configuration Provider 2.4.0 Changelog
Features Added
- Added startup retry, if the initial load fails, the provider will retry until the startup timeout is reached. By default the retry period is 100s, and can be configured via the
startup_timeoutkwarg on theloadmethod. - Adds support for adding
audienceto the kwargs forloadallowing it to specify the audience for the request. - Added support for snapshot references. Configuration settings that reference a snapshot are automatically resolved and expanded during
load.
Bugs Fixed
- Fixes a bug where
feature_flag_selectscould be passed in asNonewhich resulted in an exception on load, doing this now results in loading the default feature flags. - Fixes a bug where
feature_flag_selectscouldn’t load snapshots.
Azure Content Understanding in Foundry Tools 1.0.0 Changelog
Features Added
- GA release of Azure AI Content Understanding client library for Python
- Each
ContentFieldsubclass (e.g.,StringField,NumberField) now exposes avalueproperty with a type appropriate to that subclass (e.g.,strforStringField,floatforNumberField)
Azure Content Understanding in Foundry Tools 1.0.1 Changelog
Bugs Fixed
- Removed
_models.pyistub file that caused type checkers (pyright, mypy) to only resolve 10 of 51 model classes, hiding types likeAnalysisResultandAnalyzerDefinition. The.valueproperty type information is now provided viaTYPE_CHECKINGclass redeclarations inmodels/_patch.py.
Confidential Ledger 2.0.0b2 Changelog
Bugs Fixed
- Fixed authentication failure on HTTP redirects by preserving sensitive headers during service-managed redirects within the Confidential Ledger endpoint.
Core - Client - Core 1.38.2 Changelog
Bugs Fixed
- Fixed
PipelineClient.format_urlto preserve the leading slash when the URL template starts with/?. #45218
Core - Client - Core 1.38.3 Changelog
Bugs Fixed
- Fixed
PipelineClient.format_urlto preserve trailing slash in the base URL when the URL template is query-string-only (e.g.,?key=value). #45365 - Fixed
SensitiveHeaderCleanupPolicyto persist theinsecure_domain_changeflag across retries after a cross-domain redirect. #45518
Cosmos DB 4.15.0 Changelog
Features Added
- GA support of Per Partition Automatic Failover and AvailabilityStrategy features.
Bugs Fixed
- Fixed bug where sdk was encountering a timeout issue caused by infinite recursion during the 410 (Gone) error. See PR 44770
- Fixed crash in sync and async clients when
force_refresh_on_startupwas set toNone, which could surface asAttributeError: 'NoneType' object has no attribute '_WritableLocations'during region discovery whendatabase_accountwasNone. See PR 44987 - Fixed bug where unavailable regional endpoints were dropped from the routing list instead of being kept as fallback options. See PR 45200
Identity 1.25.3 Changelog
Bugs Fixed
- Fixed an issue where an expired token could skip refresh when a recent token request was made, due to the retry delay taking precedence over expiration. (#45496)
Mixed Reality Authentication 1.0.0b2 Changelog
Other Changes
- Python 2.7 is no longer supported. Please use Python version 3.7 or later.
- This package has been deprecated and will no longer be maintained after 10-01-2025. This package will only receive security fixes until 10-01-2025.
Resource Management - Elastic SAN 2.0.0 Changelog
Resource Management - Artifactsigning 1.0.0 Changelog
Resource Management - Batch 19.0.0 Changelog
Resource Management - Commerce 6.1.0b2 Changelog
Resource Management - Communication 3.0.0b1 Changelog
Resource Management - Computebulkactions 1.0.0b1 Changelog
Resource Management - Connected Cache 1.0.0b2 Changelog
Resource Management - Data Factory 9.3.0 Changelog
Resource Management - Disconnectedoperations 1.0.0 Changelog
Resource Management - Managedops 1.0.0b1 Changelog
Resource Management - Managedops 1.0.0b2 Changelog
Resource Management - Mixed Reality 1.1.0b2 Changelog
Resource Management - NetApp Files 15.0.0 Changelog
Resource Management - Resource-Policy 1.0.0b2 Changelog
Resource Management - Service Fabric Managed Clusters 3.0.0 Changelog
Text Analytics 5.4.0 Changelog
Breaking Changes
- Changed the continuation token format. Continuation tokens generated by previous versions of azure-core are not compatible with this version.
unknown 1.0.0b13 Changelog
Other Changes
- Upgraded azure-ai-projects version.
unknown 1.0.0b14 Changelog
Other Changes
- Pin opentelemetry-semantic-conventions-ai==0.4.13.
Latest Releases
View all the latest versions of Python packages here.
Installation Instructions
To install the latest beta version of the packages, copy and paste the following commands into a terminal:
$> pip install azure-ai-agentserver-core==1.0.0b15
$> pip install azure-ai-agentserver-core==1.0.0b16
$> pip install azure-ai-agentserver-core==1.0.0b13
$> pip install azure-ai-agentserver-core==1.0.0b14
$> pip install azure-ai-contentunderstanding==1.0.0
$> pip install azure-ai-contentunderstanding==1.0.1
$> pip install azure-ai-evaluation==1.15.1
$> pip install azure-ai-evaluation==1.15.2
$> pip install azure-ai-evaluation==1.15.3
$> pip install azure-ai-evaluation==1.16.0
$> pip install azure-ai-projects==2.0.0b4
$> pip install azure-ai-projects==2.0.0
$> pip install azure-ai-projects==2.0.1
$> pip install azure-ai-textanalytics==5.4.0
$> pip install azure-appconfiguration-provider==2.4.0
$> pip install azure-confidentialledger==2.0.0b2
$> pip install azure-core==1.38.2
$> pip install azure-core==1.38.3
$> pip install azure-cosmos==4.15.0
$> pip install azure-identity==1.25.3
$> pip install azure-mgmt-artifactsigning==1.0.0
$> pip install azure-mgmt-batch==19.0.0
$> pip install azure-mgmt-commerce==6.1.0b2
$> pip install azure-mgmt-communication==3.0.0b1
$> pip install azure-mgmt-computebulkactions==1.0.0b1
$> pip install azure-mgmt-connectedcache==1.0.0b2
$> pip install azure-mgmt-datafactory==9.3.0
$> pip install azure-mgmt-disconnectedoperations==1.0.0
$> pip install azure-mgmt-elasticsan==2.0.0
$> pip install azure-mgmt-managedops==1.0.0b1
$> pip install azure-mgmt-managedops==1.0.0b2
$> pip install azure-mgmt-mixedreality==1.1.0b2
$> pip install azure-mgmt-netapp==15.0.0
$> pip install azure-mgmt-resource-policy==1.0.0b2
$> pip install azure-mgmt-servicefabricmanagedclusters==3.0.0
$> pip install azure-mixedreality-authentication==1.0.0b2
Feedback
If you have a bug or feature request for one of the libraries, please post an issue to GitHub.