{"id":1305,"date":"2025-10-07T17:58:44","date_gmt":"2025-10-08T00:58:44","guid":{"rendered":"https:\/\/devblogs.microsoft.com\/foundry\/?p=1305"},"modified":"2025-10-09T12:52:56","modified_gmt":"2025-10-09T19:52:56","slug":"whats-new-in-azure-ai-foundry-september-2025","status":"publish","type":"post","link":"https:\/\/devblogs.microsoft.com\/foundry\/whats-new-in-azure-ai-foundry-september-2025\/","title":{"rendered":"What&#8217;s new in Azure AI Foundry | September 2025"},"content":{"rendered":"<h2>TL;DR<\/h2>\n<ul>\n<li><strong>Models (GA + Preview):<\/strong> GPT\u20115\u2011Codex (GA) brings multimodal code reasoning &amp; repo intelligence; Sora adds video\u2011to\u2011video (Preview); Grok 4 Fast (Preview) enables low\u2011latency reasoning \/ non\u2011reasoning routing; o4\u2011mini RFT graduates to GA for reward\u2011driven optimization.<\/li>\n<li><strong>Realtime &amp; Voice:<\/strong> gpt\u2011realtime (GA) unifies low\u2011latency speech \u2194 reasoning; Voice Live API GA + new Marin\/Cedar voices; Avatar 4K + Emotion delivers higher\u2011fidelity expressive agents.<\/li>\n<li><strong>Agents &amp; Tools:<\/strong> Browser Automation (Preview) gains resilient DOM + telemetry; new Computer Use (Preview) unlocks pixel workflows; Python &amp; .NET Agents betas add multi\u2011tool helper APIs, streaming tool deltas, and AOT readiness.<\/li>\n<li><strong>Retrieval &amp; Evaluation:<\/strong> Azure AI Search knowledge sources + answer synthesis (Preview) replace single target index patterns; <code>azure-ai-evaluation<\/code> 1.11.0 adds experiment tags, groundedness-with-tool-context, and multilingual red team categories.<\/li>\n<li><strong>Platform &amp; Security:<\/strong> Key Vault connections (New) eliminate inline secrets; Identity SDKs (all major languages) enforce required env vars + clearer claims challenges; structured outputs &amp; function calling converge on \/openai\/v1.<\/li>\n<li><strong>Language &amp; Data Services:<\/strong> Translator API (Public Preview) integrates real\u2011time + batch multilingual flows; Text Analytics 6.0.0b1 debuts next\u2011gen surface; Foundry Local v0.7 expands Windows NPU coverage.<\/li>\n<\/ul>\n<h2>Models<\/h2>\n<h3>GPT-5-Codex (GA)<\/h3>\n<p>OpenAI\u2019s GPT\u20115\u2011Codex is now generally available in Azure AI Foundry. It unifies multimodal code reasoning (text + image context) with advanced repository awareness:<\/p>\n<ul>\n<li>Multimodal coding: Provide architecture diagrams or UI screenshots alongside code \/ build logs for contextually coherent fixes.<\/li>\n<li>Deep refactors &amp; migration flows: Long\u2011context reasoning supports staged changes (IaC, build scripts, tests) with function calling to execute or validate steps.<\/li>\n<li>Code review intelligence: Trained to surface critical dependency, security, and performance patterns early\u2014reducing manual review cycles.\nAccess &amp; deployment: Locate GPT\u20115\u2011Codex in the Model Catalog (Standard Global). Use the v1 latest REST \/ SDK surface; avoid legacy preview API parameters unless consuming preview\u2011only capabilities. Pricing table (see announcement) includes cached input tier\u2014incorporate <code>cache_control<\/code> only where reuse materially lowers token spend.\nMigration guidance: Prior Codex \/ codex\u2011mini workflows can usually migrate by (1) re\u2011deploying under a new deployment name (treat that name as <code>model<\/code>), (2) re\u2011enabling tool\/function calling with structured JSON schema, and (3) validating temperature &amp; reasoning tokens (if customizing). Evaluate hallucination deltas with groundedness metrics in <code>azure-ai-evaluation<\/code> 1.11.0.<\/li>\n<\/ul>\n<p><iframe width=\"560\" height=\"315\" src=\"https:\/\/www.youtube.com\/embed\/TEw4zDrqWfA?si=RfBpbA5a4OCnmDrY\" allowfullscreen><\/iframe><\/p>\n<div class=\"d-flex\"><a class=\"cta_button_link btn-secondary\" href=\"https:\/\/techcommunity.microsoft.com\/blog\/azure-ai-foundry-blog\/announcing-gpt%E2%80%915%E2%80%91codex-redefining-developer-experience-in-azure-ai-foundry\/4455524\" target=\"_blank\" rel=\"noopener\">Explore\u00a0GPT\u20115\u2011Codex\u00a0announcement<\/a><\/div>\n<h3>Sora video-to-video (Preview)<\/h3>\n<p>Sora expands beyond image\u2011to\u2011video into early video\u2011to\u2011video transformation: supply an input clip plus natural language directives (\u201cretain scene pacing; stylize lighting cyber\u2011noir; smooth camera jitter\u201d). Output preserves temporal coherence while applying style \/ minor layout edits. Current preview guardrails: capped clip duration, resolution ceilings, stricter content safety (strong rejection for unapproved sensitive categories). Apply via existing Sora preview access form if not already onboarded. Use for: creative iteration loops, marketing variant generation, synthetic dataset augmentation (post review of policy). Not for: photorealistic biometric reenactment or disallowed content categories.<\/p>\n<div class=\"d-flex\"><a class=\"cta_button_link btn-secondary\" href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/openai\/video-generation-quickstart?tabs=windows%2Ckeyless\" target=\"_blank\" rel=\"noopener\">Try\u00a0Sora\u00a0video\u00a0quickstart<\/a><\/div>\n<h3>gpt\u2011realtime (GA)<\/h3>\n<p>The new gpt\u2011realtime model unifies Azure\u2019s speech\u2011to\u2011speech capabilities\u2014improved instruction following, lower latency, higher naturalness, and image input support. It collapses prior multi\u2011model pipelines (STT \u2192 LLM \u2192 TTS) into a single low\u2011latency loop, lowering end\u2011to\u2011end round\u2011trip times for live support, IVR deflection, education, and collaborative coding voice agents. New Marin &amp; Cedar voices add expressive prosody; pricing undercuts the earlier realtime preview by ~20%. Combine with Voice Live API session state for resilient interruption handling and with function calling to execute quick lookups or transactions mid\u2011dialog.<\/p>\n<p>Key advantages:<\/p>\n<ul>\n<li>Fewer moving parts (simpler scaling &amp; observability)<\/li>\n<li>Better barge\u2011in \/ interruption handling<\/li>\n<li>Native tool\/function calling for orchestration<\/li>\n<li>Direct multimodal (describe image via voice, then refine)<\/li>\n<\/ul>\n<div class=\"d-flex\"><a class=\"cta_button_link btn-secondary\" href=\"https:\/\/techcommunity.microsoft.com\/blog\/azure-ai-foundry-blog\/announcing-gpt-realtime-on-azure-ai-foundry\/4449666\" target=\"_blank\" rel=\"noopener\">Read\u00a0gpt\u2011realtime\u00a0GA\u00a0blog<\/a><\/div>\n<h3>Grok 4 Fast models (Preview)<\/h3>\n<p><code>grok-4-fast-reasoning<\/code> and <code>grok-4-fast-non-reasoning<\/code> from xAI bring parallel function calling, long (~131K) context, multimodal image understanding (with tokenizer), and structured JSON output schemas tuned for agentic tool orchestration. Choose the reasoning variant for complex planning \/ analytical hops; choose non\u2011reasoning for deterministic low\u2011latency classification, extraction, or lightweight chaining. Efficient H100 utilization lowers per\u2011request latency, allowing cost\/quality routing (e.g., default to non\u2011reasoning; escalate to reasoning when confidence or complexity thresholds are unmet).<\/p>\n<p>Implementation tips:<\/p>\n<ul>\n<li>Define JSON Schema for predictable tool outputs\u2014minimizes downstream validation.<\/li>\n<li>Log reasoning depth metrics (tokens\/time) to refine router thresholds.<\/li>\n<li>Pair with answer synthesis evaluation to measure citation correctness.<\/li>\n<\/ul>\n<div class=\"d-flex\"><a class=\"cta_button_link btn-secondary\" href=\"https:\/\/techcommunity.microsoft.com\/blog\/azure-ai-foundry-blog\/announcing-the-grok-4-fast-models-from-xai-now-available-in-azure-ai-foundry\/4456701\" target=\"_blank\" rel=\"noopener\">Preview\u00a0Grok\u00a04\u00a0Fast\u00a0models<\/a><\/div>\n<h3>o4\u2011mini Reinforcement Fine-Tuning (GA)<\/h3>\n<p>Reinforcement Fine-Tuning (RFT) for <code>o4-mini<\/code> graduates to GA. Instead of curating thousands of labeled completions (SFT), supply a grading function or choose built\u2011in graders (code correctness, groundedness, format adherence, safety) and let iterative sampling optimize for reward. Start effective runs with ~100 scored examples; expand only if reward plateaus. Use RFT when: multi\u2011step reasoning consistency matters; policy adherence needs dynamic evaluation; or SFT plateaued on edge\u2011case reasoning.<\/p>\n<p>Workflow outline:<\/p>\n<ol>\n<li>Draft concise rubric (YAML \/ Python grader) with explicit pass\/fail + optional scalar score.<\/li>\n<li>Seed minimal diverse prompts (avoid near duplicates).<\/li>\n<li>Kick off RFT job; monitor reward trend &amp; variance.<\/li>\n<li>Post\u2011train, re\u2011run evaluation suite (groundedness, red team) to validate safety\/regression.<\/li>\n<\/ol>\n<p><img decoding=\"async\" src=\"https:\/\/devblogs.microsoft.com\/foundry\/wp-content\/uploads\/sites\/89\/2025\/10\/o4mini_Python.gif\" alt=\"o4-mini RFT fine-tuning Python grading loop example showing iterative reward improvements\" \/><\/p>\n<div class=\"d-flex\"><a class=\"cta_button_link btn-secondary\" href=\"https:\/\/techcommunity.microsoft.com\/blog\/azure-ai-foundry-blog\/o4-mini-reinforcement-fine-tuning-rft-now-generally-available-on-azure-ai-foundr\/4452597\" target=\"_blank\" rel=\"noopener\">Learn\u00a0about\u00a0o4\u2011mini\u00a0RFT\u00a0GA<\/a><\/div>\n<div>\n<h3>Local + edge<\/h3>\n<ul>\n<li>Foundry Local v0.7: Expanded NPU support (Intel &amp; AMD) on Windows 11, dynamic execution provider detection via Windows ML\u2014simplifies distributing hybrid local\/cloud fallback apps.<\/li>\n<\/ul>\n<div class=\"d-flex\" style=\"gap: .5rem; flex-wrap: wrap;\"><a class=\"cta_button_link btn-secondary\" href=\"https:\/\/techcommunity.microsoft.com\/blog\/azure-ai-foundry-blog\/upgrade-your-voice-agent-with-azure-ai-voice-live-api\/4458247\" target=\"_blank\" rel=\"noopener\">Voice\u00a0Live\u00a0API\u00a0GA<\/a>\n<a class=\"cta_button_link btn-secondary\" href=\"http:\/\/aka.ms\/foundry-local-docs\" target=\"_blank\" rel=\"noopener\">Foundry\u00a0Local\u00a0docs<\/a><\/div>\n<\/div>\n<h3>Avatar 4K + Emotion Control<\/h3>\n<p>High\u2011fidelity 4K avatar rendering plus synchronized emotion control (voice + facial expressions) elevate customer support, training, accessibility, and marketing use cases. Train with 4K source video (framed &amp; centered) for maximum detail retention; choose portrait for vertical signage \/ mobile, landscape for presentations. Emotion parameters (e.g., cheerful, angry, hopeful) propagate to both prosody and facial animation, producing consistent affective output\u2014falling back to neutral where unsupported.<\/p>\n<p>Best practices:<\/p>\n<ul>\n<li>Downscale at runtime (e.g., 1920\u00d71080) if bandwidth is constrained\u2014retain original 4K master for archival.<\/li>\n<li>Use SSML emotion tags for fine\u2011grained segment control within longer scripts.<\/li>\n<li>Benchmark MOS (expressiveness) for your domain before large rollout.<\/li>\n<\/ul>\n<p><img decoding=\"async\" src=\"https:\/\/devblogs.microsoft.com\/foundry\/wp-content\/uploads\/sites\/89\/2025\/10\/speech-playground.png\" alt=\"Avatar emotion control preview in speech playground showing selectable emotion parameters and synchronized facial expression rendering\" \/><\/p>\n<div class=\"d-flex\"><a class=\"cta_button_link btn-secondary\" href=\"https:\/\/techcommunity.microsoft.com\/blog\/azure-ai-foundry-blog\/latest-avatar-updates-4k-support-and-emotion-control-features\/4449995\" target=\"_blank\" rel=\"noopener\">See\u00a0Avatar\u00a04K\u00a0+\u00a0Emotion\u00a0features<\/a><\/div>\n<h2>Agents<\/h2>\n<h3>Computer Use tool (Preview)<\/h3>\n<p>Enables pixel-level interaction (click, type, screenshot feedback loop) across desktop or browser surfaces when DOM semantics are insufficient (legacy apps, canvas\u2011heavy UIs). Compared to Browser Automation, Computer Use trades semantic DOM awareness for raw visual adaptability\u2014powerful but higher risk.<\/p>\n<p>Risk mitigations:<\/p>\n<ul>\n<li>Run only in isolated, low\u2011privilege VMs without production secrets.<\/li>\n<li>Enforce human approval for domain \/ action categories (file system, administrative panels).<\/li>\n<li>Maintain an allowlist of permitted host\/application windows; reject off\u2011scope UI transitions.<\/li>\n<\/ul>\n<p>Telemetry tip: Persist screenshot hashes to detect visually static loops (stalled automation) and abort gracefully.<\/p>\n<div class=\"d-flex\"><a class=\"cta_button_link btn-secondary\" href=\"https:\/\/devblogs.microsoft.com\/foundry\/announcing-computer-use-tool-preview-in-azure-ai-foundry-agent-service\/\" target=\"_blank\" rel=\"noopener\">Computer\u00a0Use\u00a0tool\u00a0preview<\/a><\/div>\n<h3>Microsoft Agent Framework: the new OSS orchestration engine for agent development<\/h3>\n<p>The Microsoft Agent Framework (OSS) unifies Semantic Kernel durability &amp; AutoGen orchestration. Early adopters benefit from: MCP discovery, agent\u2011to\u2011agent (A2A) messaging, declarative YAML\/JSON agent specs, OpenTelemetry tracing, and enterprise approval hooks. Plan migration: map legacy plugin patterns to Tool abstractions, validate memory connectors, re\u2011instrument tracing IDs for continuity in dashboards.<\/p>\n<p><iframe width=\"560\" height=\"315\" src=\"https:\/\/www.youtube.com\/embed\/yOBcPuLLmuY?si=b7EGjnnreCwoAacY\" allowfullscreen><\/iframe><\/p>\n<div class=\"d-flex\"><a class=\"cta_button_link btn-secondary\" href=\"https:\/\/aka.ms\/AgentFramework\" target=\"_blank\" rel=\"noopener\">Explore\u00a0Microsoft\u00a0Agent\u00a0Framework<\/a><\/div>\n<h2>Tools<\/h2>\n<h3>Evaluation &amp; safety<\/h3>\n<p><code>azure-ai-evaluation<\/code> 1.11.0 introduces:<\/p>\n<ul>\n<li>Experiment tags: Attach stable metadata (model deployment name, router policy hash, tool set version) for slice comparisons.<\/li>\n<li>Groundedness with tool context: Evaluator now inspects cited tool outputs and retrieved documents\u2014flagging hallucinations that conflict with authoritative snippets.<\/li>\n<li>Multilingual red teaming expansion: Adds IndirectAttack (prompt injection \/ covert instruction) &amp; UngroundedAttributes categories for broader safety coverage.<\/li>\n<\/ul>\n<p>Recommended workflow: (1) Tag baseline (pre\u2011Browser Automation) agent, (2) enable new tools, (3) re\u2011run evaluation suite; compare groundedness deltas and risk incidence to justify rollout.<\/p>\n<div class=\"d-flex\"><a class=\"cta_button_link btn-secondary\" href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/concepts\/observability\" target=\"_blank\" rel=\"noopener\">Review\u00a0evaluation\u00a0evaluators<\/a><\/div>\n<h3>Retrieval (Azure AI Search)<\/h3>\n<p>Knowledge sources (Preview) abstract underlying indexes &amp; Blob sources with integrated vectorization + chunking. Answer synthesis composes grounded responses with inline citations across multiple knowledge sources\u2014removing earlier single target index constraints. Combine with index description metadata to improve relevance selection. Debug hybrid retrieval via sub\u2011score &amp; debug modes; apply scoring profiles during semantic reranking for domain boosting (freshness, popularity, compliance tags).<\/p>\n<div class=\"d-flex\"><a class=\"cta_button_link btn-secondary\" href=\"https:\/\/techcommunity.microsoft.com\/blog\/azure-ai-foundry-blog\/agentic-retrieval-updates-in-azure-ai-search\/4450621\" target=\"_blank\" rel=\"noopener\">Agentic retrieval updates<\/a><\/div>\n<h2>Platform<\/h2>\n<h3>Translator API<\/h3>\n<p>Add neural machine translation, custom terminology, and document translation pipelines directly into multi\u2011modal or agentic workflows. Use Translator alongside GPT\u20115\u2011Codex or retrieval flows to localize generated content, summarize foreign\u2011language sources, or produce parallel corpora for evaluation.<\/p>\n<p>Key scenarios:<\/p>\n<ul>\n<li>Real\u2011time chat + translation: route user utterances through Translator before\/after model reasoning.<\/li>\n<li>Batch document translation: feed translated artifacts into knowledge sources for multi\u2011lingual grounding.<\/li>\n<li>Terminology compliance: apply custom term lists prior to evaluation to reduce post\u2011edit distance.<\/li>\n<\/ul>\n<p><iframe width=\"560\" height=\"315\" src=\"https:\/\/www.youtube.com\/embed\/Xam_QQnb5wQ?si=7R1bnpVMBCrus1J3\" allowfullscreen><\/iframe><\/p>\n<h2>SDK &amp; language changelog (September 2025)<\/h2>\n<p>Focused highlights for the core Azure AI Foundry developer packages this month. Each bullet groups all September versions (stable + beta) so you can scan breaking changes, new features, and migration actions quickly.<\/p>\n<h3>Python<\/h3>\n<p><strong>Agents (<code>azure-ai-agents<\/code> 1.2.0b2 \u2192 1.2.0b4)<\/strong>\nFeatures: Browser Automation tool support; Computer Use preview tool; multi\u2011tool helper APIs (<code>get_tool_resources<\/code>, <code>get_tool_definitions<\/code>); streaming delta models for Connected \/ search \/ Fabric \/ SharePoint tools.\nReliability: Multiple tracing fixes (duplicate span name normalization, hanging <code>create_and_process<\/code> w\/ MCP approval).\nAction: If you instrument OpenTelemetry, re-run baselines\u2014span naming updates may affect existing dashboards.\n<a href=\"https:\/\/github.com\/Azure\/azure-sdk-for-python\/tree\/azure-ai-agents_1.2.0b4\/sdk\/ai\/azure-ai-agents\/CHANGELOG.md\">Changelog<\/a><\/p>\n<p><strong>Evaluation (<code>azure-ai-evaluation<\/code> 1.11.0)<\/strong>\nFeatures: Experiment <code>tags<\/code>, groundedness evaluator extended to tool outputs (file search context), multilingual red team (IndirectAttack, UngroundedAttributes).\nAction: Start tagging evaluation runs (e.g., model deployment, tool policy hash) to enable cross\u2011version diff queries.\n<a href=\"https:\/\/github.com\/Azure\/azure-sdk-for-python\/tree\/azure-ai-evaluation_1.11.0\/sdk\/evaluation\/azure-ai-evaluation\/CHANGELOG.md\">Changelog<\/a><\/p>\n<p><strong>Projects (<code>azure-ai-projects<\/code> 1.1.0b3\u2013b4)<\/strong>\nChanges: Dependency bump to prefer the newer Agents beta (<code>&gt;=1.2.0b3<\/code>), secret key retrieval bug fix for custom key connections.\nAction: Pin <code>azure-ai-agents&gt;=1.2.0b4<\/code> to pick up tracing + tool improvements before scaling multi\u2011tool threads.\n<a href=\"https:\/\/github.com\/Azure\/azure-sdk-for-python\/tree\/azure-ai-projects_1.1.0b4\/sdk\/ai\/azure-ai-projects\/CHANGELOG.md\">Changelog<\/a><\/p>\n<p><strong>Search (<code>azure-search-documents<\/code> 11.7.0b1)<\/strong>\nFeatures: Knowledge sources abstraction, answer synthesis, extensive new model classes (KnowledgeSource*, agent activity records), removal of target index constructs.\nBreaking (preview): Removal of <code>KnowledgeAgentTargetIndex<\/code> &amp; related agent index params\u2014migrate to <code>KnowledgeSource<\/code> objects.\n<a href=\"https:\/\/github.com\/Azure\/azure-sdk-for-python\/tree\/azure-search-documents_11.7.0b1\/sdk\/search\/azure-search-documents\/CHANGELOG.md\">Changelog<\/a><\/p>\n<p><strong>Identity (<code>azure-identity<\/code> 1.25.0)<\/strong>\nFeatures: <code>claims<\/code> support for <code>AzureDeveloperCliCredential<\/code>; <code>require_envvar<\/code> to enforce env presence; multiple auth robustness fixes.\nAction: Use <code>DefaultAzureCredential(require_envvar=\"AZURE_TOKEN_CREDENTIALS\")<\/code> in CI to fail fast on misconfiguration.\n<a href=\"https:\/\/github.com\/Azure\/azure-sdk-for-python\/tree\/azure-identity_1.25.0\/sdk\/identity\/azure-identity\/CHANGELOG.md\">Changelog<\/a><\/p>\n<p><strong>Text Analytics (<code>azure-ai-textanalytics<\/code> 6.0.0b1)<\/strong>\nStatus: First beta of next\u2011gen surface; expect breaking shape evolution\u2014keep isolated virtual env during evaluation.\n<a href=\"https:\/\/github.com\/Azure\/azure-sdk-for-python\/tree\/azure-ai-textanalytics_6.0.0b1\/sdk\/textanalytics\/azure-ai-textanalytics\/CHANGELOG.md\">Changelog<\/a><\/p>\n<p><strong>Voice Live (<code>azure-ai-voicelive<\/code> 1.0.0b1\u2013b2)<\/strong>\nFeatures: WebSocket connection model (<code>VoiceLiveConnection<\/code>), async function call support, serialization fix for function outputs.\nAction: When upgrading from b1 \u2192 b2, validate function call message schema (JSON str serialization change).\n<a href=\"https:\/\/github.com\/Azure\/azure-sdk-for-python\/tree\/azure-ai-voicelive_1.0.0b2\/sdk\/ai\/azure-ai-voicelive\/CHANGELOG.md\">Changelog (b2)<\/a><\/p>\n<p><strong>Cosmos DB (<code>azure-cosmos<\/code> 4.14.0b2\u2013b4)<\/strong>\nFocus: Session token scoping optimizations, extensive bug fixes (health check region handling, partition key edge cases), new feed range &amp; multi\u2011item read APIs.\nAction: Re\u2011benchmark RU &amp; latency after upgrade\u2014per request session token trimming can materially reduce header size.\n<a href=\"https:\/\/github.com\/Azure\/azure-sdk-for-python\/tree\/azure-cosmos_4.14.0b4\/sdk\/cosmos\/azure-cosmos\/CHANGELOG.md\">Latest beta changelog<\/a><\/p>\n<h3>.NET<\/h3>\n<p><strong>Agents Persistent (<code>Azure.AI.Agents.Persistent<\/code> 1.2.0-beta.2\u2013.5)<\/strong>\nFeatures: Streaming MCP tool support, FileSearch &amp; CodeInterpreter tool enablement, AOT compatibility, additional delta tool call models, delete ThreadMessages, Bing\/Fabric\/SharePoint tool deltas.\nFixes: ResponseFormat handling + unexpected streaming type issues, image block serialization.\nAction: If targeting native AOT, update to \u2265 beta.4 and re\u2011generate RDXML after adding new tool types.\n<a href=\"https:\/\/github.com\/Azure\/azure-sdk-for-net\/tree\/Azure.AI.Agents.Persistent_1.2.0-beta.5\/sdk\/ai\/Azure.AI.Agents.Persistent\/CHANGELOG.md\">Changelog (latest)<\/a><\/p>\n<p><strong>Projects (<code>Azure.AI.Projects<\/code> 1.0.0-beta.11)<\/strong>\nBreaking: Extensive type and method renames (e.g., <code>AIDeployment<\/code> \u2192 <code>AssetDeployment<\/code>, consolidated <code>GetOpenAIClient<\/code>).\nFeatures: New <code>AIProjectClient<\/code> ctor supporting <code>AuthenticationTokenProvider<\/code>.\nAction: Run a full solution rename pass; integrate with unified Inference client path.\n<a href=\"https:\/\/github.com\/Azure\/azure-sdk-for-net\/tree\/Azure.AI.Projects_1.0.0-beta.11\/sdk\/ai\/Azure.AI.Projects\/CHANGELOG.md\">Changelog<\/a><\/p>\n<p><strong>OpenAI Inference (<code>Azure.AI.OpenAI<\/code> 2.3.0-beta.2)<\/strong>\nFeatures: Experimental default headers &amp; query parameter injection (pipeline customization).\nAction: Avoid embedding secrets\u2014use this for correlation IDs or feature flags only.\n<a href=\"https:\/\/github.com\/Azure\/azure-sdk-for-net\/tree\/Azure.AI.OpenAI_2.3.0-beta.2\/sdk\/openai\/Azure.AI.OpenAI\/CHANGELOG.md\">Changelog<\/a><\/p>\n<p><strong>Search (<code>Azure.Search.Documents<\/code> 11.7.0-beta.6\u2013.7)<\/strong>\nFeatures: System.ClientModel serialization; knowledge sources &amp; answer synthesis; AOT\u2011friendly overloads with <code>JsonTypeInfo&lt;T&gt;<\/code>.\nBreaking: Removal of knowledge agent target index; <code>QueryDebugMode<\/code> relocation.\n<a href=\"https:\/\/github.com\/Azure\/azure-sdk-for-net\/tree\/Azure.Search.Documents_11.7.0-beta.7\/sdk\/search\/Azure.Search.Documents\/CHANGELOG.md\">Changelog (beta.7)<\/a><\/p>\n<p><strong>Identity (<code>Azure.Identity<\/code> 1.16.0 + <code>Azure.Identity.Broker<\/code> 1.3.0)<\/strong>\nFeatures: Custom env variable constructor, improved claims challenge exceptions, reduced probe traffic with Managed Identity env configuration; broker package deprecates legacy SharedTokenCache credential options.\nFixes: <code>AzureDeveloperCliCredential<\/code> hang &amp; token cache issues.\n<a href=\"https:\/\/github.com\/Azure\/azure-sdk-for-net\/tree\/Azure.Identity_1.16.0\/sdk\/identity\/Azure.Identity\/CHANGELOG.md\">Changelog<\/a><\/p>\n<p><strong>Core (<code>Azure.Core<\/code> 1.47.2\u20131.48.0 \/ <code>System.ClientModel<\/code> 1.6.0\u20131.6.1)<\/strong>\nUpdates: Adoption of System.ClientModel 1.6.x, user agent policy additions, AOT friendliness improvements, auth bug fix enabling TokenCredential with unbranded clients.\n<a href=\"https:\/\/github.com\/Azure\/azure-sdk-for-net\/tree\/Azure.Core_1.48.0\/sdk\/core\/Azure.Core\/CHANGELOG.md\">Azure.Core changelog<\/a><\/p>\n<p><strong>Data Movement (Storage) 12.2.2 &amp; 12.3.0-beta.1<\/strong>\nFixes: Path handling for directory uploads; metadata preservation.\nFeatures (beta): Quality &amp; cross-protocol copy improvements; SMB\u2194NFS share copy support (files shares beta).\n<a href=\"https:\/\/github.com\/Azure\/azure-sdk-for-net\/tree\/Azure.Storage.DataMovement_12.3.0-beta.1\/sdk\/storage\/Azure.Storage.DataMovement\/CHANGELOG.md\">Changelog<\/a><\/p>\n<p><strong>Synapse Artifacts 1.0.0-preview.22<\/strong>\nLarge schema expansion (ServiceNowV2, SnowflakeV2, cert-based auth params, numerous Oracle \/ Amazon RDS additions) plus redaction\/bypass flags.\nAction: Regenerate strongly typed orchestration layers if you rely on generated linked service definitions.\n<a href=\"https:\/\/github.com\/Azure\/azure-sdk-for-net\/tree\/Azure.Analytics.Synapse.Artifacts_1.0.0-preview.22\/sdk\/synapse\/Azure.Analytics.Synapse.Artifacts\/CHANGELOG.md\">Changelog<\/a><\/p>\n<h3>Java<\/h3>\n<p><strong>Search (<code>azure-search-documents<\/code> 11.7.9 GA &amp; 11.8.0-beta.8)<\/strong>\nBeta: Knowledge sources + answer synthesis, dropped obsolete target index constructs, added <code>VectorFilterMode.STRICT_POST_FILTER<\/code>.\nGA patch: Dependency upgrades (<code>azure-core<\/code> 1.56.0, serializer updates).\n<a href=\"https:\/\/github.com\/Azure\/azure-sdk-for-java\/tree\/azure-search-documents_11.8.0-beta.8\/sdk\/search\/azure-search-documents\/CHANGELOG.md\">Beta changelog<\/a><\/p>\n<p><strong>Identity (<code>azure-identity<\/code> 1.18.0 + <code>azure-identity-broker<\/code> 1.1.17)<\/strong>\nFeatures: Claims challenge support across CLI\/PowerShell credentials; <code>AzureIdentityEnvVars<\/code> enum + <code>requireEnvVars<\/code> builder enforcement; enhanced error messaging.\n<a href=\"https:\/\/github.com\/Azure\/azure-sdk-for-java\/tree\/azure-identity_1.18.0\/sdk\/identity\/azure-identity\/CHANGELOG.md\">Changelog<\/a><\/p>\n<p><strong>Cosmos (<code>azure-cosmos<\/code> 4.74.0 &amp; ecosystem)<\/strong>\nAdds throughput bucket support; encryption &amp; Spark connectors updated; test harness beta adds Gateway V2.\nAction: Evaluate new Spark auth audience config (<code>spark.cosmos.auth.aad.audience<\/code>).\n<a href=\"https:\/\/github.com\/Azure\/azure-sdk-for-java\/tree\/azure-cosmos_4.74.0\/sdk\/cosmos\/azure-cosmos\/CHANGELOG.md\">Changelog<\/a><\/p>\n<p><strong>Communication Call Automation 1.4.3 \/ 1.5.0 \/ 1.6.0-beta.1<\/strong>\nProgressive feature rollout: Teams persona support, public participant transfer (beta), transcription PII redaction &amp; sentiment analysis.\nAction: Adopt 1.5.0 for stable recording improvements; experiment with beta for advanced transfer flows.\n<a href=\"https:\/\/github.com\/Azure\/azure-sdk-for-java\/tree\/azure-communication-callautomation_1.6.0-beta.1\/sdk\/communication\/azure-communication-callautomation\/CHANGELOG.md\">Changelog (1.6.0-beta.1)<\/a><\/p>\n<p><strong>Storage (Blobs 12.31.2 + batch\/crypto)<\/strong>\nDependency alignment with <code>azure-core<\/code> 1.56.0; cryptography &amp; internal Avro updates\u2014reassess shaded deps if you build a fat JAR.\n<a href=\"https:\/\/github.com\/Azure\/azure-sdk-for-java\/tree\/azure-storage-blob_12.31.2\/sdk\/storage\/azure-storage-blob\/CHANGELOG.md\">Blob changelog<\/a><\/p>\n<p><strong>Core (<code>azure-core<\/code> 1.56.1 &amp; serializers)<\/strong>\nSerializer (Jackson\/Gson) bumps; ensure any custom <code>ObjectMapper<\/code> modules are compatible after upgrade.\n<a href=\"https:\/\/github.com\/Azure\/azure-sdk-for-java\/tree\/azure-core_1.56.1\/sdk\/core\/azure-core\/CHANGELOG.md\">Core changelog<\/a><\/p>\n<h3>JavaScript \/ TypeScript<\/h3>\n<p><strong>Monitor OpenTelemetry 1.13.0 \u2192 1.14.0<\/strong>\nFeatures: Enable nearly all instrumentations by default; multi\u2011exporter metrics support; default exporter version bump.\nAction: Audit duplicated custom instrumentation to avoid double spans after upgrade.\n<a href=\"https:\/\/github.com\/Azure\/azure-sdk-for-js\/tree\/@azure\/monitor-opentelemetry_1.14.0\/sdk\/monitor\/monitor-opentelemetry\/CHANGELOG.md\">Changelog (1.14.0)<\/a><\/p>\n<p><strong>Communication Call Automation 1.5.0 &amp; 1.6.0-beta.1<\/strong>\nAdds Teams multipersona, recording via call connection ID, sentiment + PII redaction (beta adds participant transfer, richer transcription tooling).\n<a href=\"https:\/\/github.com\/Azure\/azure-sdk-for-js\/tree\/@azure\/communication-call-automation_1.6.0-beta.1\/sdk\/communication\/communication-call-automation\/CHANGELOG.md\">Changelog (beta)<\/a><\/p>\n<p><strong>Identity 4.12.0<\/strong>\nFeatures: <code>requiredEnvVars<\/code> option + union type; claims challenge failure clarity; multiple credential UX fixes.\nAction: Set <code>requiredEnvVars: [\"AZURE_TOKEN_CREDENTIALS\"]<\/code> in production initialization blocks for deterministic startup validation.\n<a href=\"https:\/\/github.com\/Azure\/azure-sdk-for-js\/tree\/@azure\/identity_4.12.0\/sdk\/identity\/identity\/CHANGELOG.md\">Changelog<\/a><\/p>\n<p><strong>Cosmos (<code>@azure\/cosmos<\/code> 4.5.1)<\/strong>\nReliability: Removal of unavailable regions from location cache; retry policy refinement for ENOTFOUND.\nAction: Clear any custom location preference overrides to benefit from new pruning heuristics.\n<a href=\"https:\/\/github.com\/Azure\/azure-sdk-for-js\/tree\/@azure\/cosmos_4.5.1\/sdk\/cosmosdb\/cosmos\/CHANGELOG.md\">Changelog<\/a><\/p>\n<p><strong>Playwright ecosystem<\/strong>\n<code>@azure\/create-playwright<\/code> 1.0.0 and <code>@azure\/playwright<\/code> 1.0.0 initial releases; deprecation notices for legacy create \/ microsoft-playwright-testing packages\u2014follow migration guide to new config names (<code>createAzurePlaywrightConfig<\/code>, <code>connectTimeout<\/code>).\n<a href=\"https:\/\/github.com\/Azure\/azure-sdk-for-js\/tree\/@azure\/create-playwright_1.0.0\/sdk\/loadtesting\/create-playwright\/CHANGELOG.md\">Create Playwright changelog<\/a><\/p>\n<p><strong>OpenTelemetry Exporter 1.0.0-beta.34\u2013.35<\/strong>\nFeatures: Customer SDK Stats export interval env var; telemetry success field; dependency refresh.\n<a href=\"https:\/\/github.com\/Azure\/azure-sdk-for-js\/tree\/@azure\/monitor-opentelemetry-exporter_1.0.0-beta.35\/sdk\/monitor\/monitor-opentelemetry-exporter\/CHANGELOG.md\">Changelog (beta.35)<\/a><\/p>\n<p><strong>TypeSpec HTTP Runtime 0.3.1<\/strong>\nLogging improvements (leaner RestError output), dependency bumps.\n<a href=\"https:\/\/github.com\/Azure\/azure-sdk-for-js\/tree\/@typespec\/ts-http-runtime_0.3.1\/sdk\/core\/ts-http-runtime\/CHANGELOG.md\">Changelog<\/a><\/p>\n<h2>Documentation updates<\/h2>\n<h3>New<\/h3>\n<ul>\n<li>GPT-5-Codex GA announcement &amp; model entry (multimodal code reasoning) \u2013 migration guidance &amp; repo intelligence patterns. (<a href=\"https:\/\/techcommunity.microsoft.com\/blog\/azure-ai-foundry-blog\/announcing-gpt%E2%80%915%E2%80%91codex-redefining-developer-experience-in-azure-ai-foundry\/4455524\">announcement<\/a> \u00b7 <a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/openai\/includes\/models-azure-direct-openai\">model entry<\/a>)<\/li>\n<li>Sora video-to-video preview docs \u2013 input clip transformation, temporal coherence guardrails. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/openai\/video-generation-quickstart\">quickstart<\/a> \u00b7 <a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/openai\/concepts\/video-generation\">concept<\/a>)<\/li>\n<li>Browser Automation tool docs &amp; samples \u2013 preview setup, DOM heuristics, Playwright Workspace integration. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/agents\/how-to\/tools\/browser-automation\">how-to<\/a> \u00b7 <a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/agents\/how-to\/tools\/browser-automation-samples\">samples<\/a>)<\/li>\n<li>Key Vault connections setup \u2013 managed secret binding with rotation playbook. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/how-to\/set-up-key-vault-connection\">guide<\/a>)<\/li>\n<li>Deep Research agents tutorial \u2013 multi-hop planning + Bing search orchestration. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/agents\/how-to\/tools\/deep-research-samples\">samples<\/a>)<\/li>\n<li>Browser Automation + Computer Use security\/isolation architecture pages. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/agents\/how-to\/tools\/browser-automation\">browser automation<\/a> \u00b7 <a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/agents\/how-to\/tools\/computer-use\">computer use<\/a>)<\/li>\n<li>Retrieval knowledge sources &amp; answer synthesis \u2013 multi-index grounding &amp; latency trade-offs. (<a href=\"https:\/\/techcommunity.microsoft.com\/blog\/azure-ai-foundry-blog\/agentic-retrieval-updates-in-azure-ai-search\/4450621\">blog<\/a>)<\/li>\n<li>Evaluation graders (groundedness with tool context, multilingual red teaming) + tagging best practices. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/concepts\/evaluation-evaluators\">evaluators<\/a>)<\/li>\n<li>Network security perimeter (preview) configuration guide \u2013 perimeter join &amp; logging. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/how-to\/add-foundry-to-network-security-perimeter\">perimeter guide<\/a>)<\/li>\n<li>Deployment policy (Azure Policy) guide \u2013 restrict model families, context length, publisher controls. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/how-to\/model-deployment-policy\">model deployment policy<\/a>)<\/li>\n<li>Consolidated cost &amp; usage dashboard concept \u2013 unified spend\/health view. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/concepts\/ai-foundry-consolidated-view\">consolidated view<\/a>)<\/li>\n<li>Model router regional expansion matrix \u2013 new regions &amp; fallback guidance. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/agents\/concepts\/model-region-support\">model region support<\/a>)<\/li>\n<li>Structured outputs + function calling modernization (\/openai\/v1) \u2013 JSON schema strategies. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/openai\/how-to\/structured-outputs\">structured outputs<\/a> \u00b7 <a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/openai\/how-to\/function-calling\">function calling<\/a>)<\/li>\n<li>Azure Direct Models terminology pages (abuse monitoring &amp; limited access policy refactor). (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/openai\/concepts\/abuse-monitoring\">abuse monitoring<\/a> \u00b7 <a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/responsible-ai\/openai\/limited-access\">limited access<\/a>)<\/li>\n<li>Local tracing quickstart (instrument locally, export to portal) &amp; advanced redaction. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/how-to\/develop\/trace-application\">trace application<\/a>)<\/li>\n<li>Computer Use tool overview &amp; samples \u2013 pixel automation + high-risk mitigations. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/agents\/how-to\/tools\/computer-use\">overview<\/a> \u00b7 <a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/agents\/how-to\/tools\/computer-use-samples\">samples<\/a>)<\/li>\n<li>Hub CMK &amp; Hub RBAC dedicated articles \u2013 separated from project scope. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/concepts\/hub-encryption-keys-portal\">hub CMK<\/a> \u00b7 <a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/concepts\/hub-rbac-azure-ai-foundry\">hub RBAC<\/a>)<\/li>\n<li>BCDR for Agents section \u2013 state durability &amp; recovery. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/agents\/overview\">agents overview<\/a>)<\/li>\n<li>Voice Live data privacy + transparency notes \u2013 processing &amp; retention. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/responsible-ai\/speech-service\/voice-live\/data-privacy-security\">privacy<\/a> \u00b7 <a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/responsible-ai\/speech-service\/voice-live\/transparency-note\">transparency<\/a>)<\/li>\n<li>Consolidated model deployments guide + marketplace configuration expansion. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/foundry-models\/how-to\/create-model-deployments\">deployments guide<\/a> \u00b7 <a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/foundry-models\/how-to\/configure-marketplace\">marketplace config<\/a>)<\/li>\n<li>General platform cost management concept (broad billing + budgeting) and models cost management overhaul. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/concepts\/manage-costs\">platform costs<\/a> \u00b7 <a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/foundry-models\/how-to\/manage-costs\">models costs<\/a>)<\/li>\n<li>GPT\u20115 vs GPT\u20114.1 comparison guide \u2013 model selection heuristics. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/foundry-models\/how-to\/model-choice-guide\">comparison<\/a>)<\/li>\n<li>Codex how-to guide \u2013 CLI, VS Code, approval modes, security guardrails. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/openai\/how-to\/codex\">codex<\/a>)<\/li>\n<li>AI solution templates additions \u2013 accelerator templates. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/how-to\/develop\/ai-template-get-started\">templates<\/a>)<\/li>\n<li>Continuous evaluation (agents) region &amp; metrics guide. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/how-to\/continuous-evaluation-agents\">continuous eval<\/a>)<\/li>\n<li>Structured outputs REST\/Python includes with <code>gpt-5-codex<\/code> &amp; \/openai\/v1 adoption. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/openai\/how-to\/structured-outputs\">structured outputs<\/a>)<\/li>\n<\/ul>\n<h3>Updated \/ GA \/ Overhauled<\/h3>\n<ul>\n<li>GPT-5-Codex model docs \u2013 moved from limited access to GA; added quotas &amp; reasoning feature clarifications. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/openai\/includes\/models-azure-direct-openai\">model entry<\/a>)<\/li>\n<li>Sora docs \u2013 video-to-video workflow, multi-input constraints, added region (Sweden Central) &amp; REST examples. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/openai\/concepts\/video-generation\">concept<\/a>)<\/li>\n<li>gpt-realtime \u2013 GA rename, new event names, latency &amp; supported models list updates. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/openai\/realtime-audio-quickstart\">realtime quickstart<\/a>)<\/li>\n<li>Grok 4 \/ Grok 4 Fast models \u2013 region &amp; capability tables, availability columns added\/normalized. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/foundry-models\/includes\/models-azure-direct-others\">models include<\/a>)<\/li>\n<li>o4-mini RFT \u2013 GA guidance, benchmark methodology &amp; reward tuning recommendations. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/openai\/how-to\/reinforcement-fine-tuning\">RFT how-to<\/a>)<\/li>\n<li>Avatar 4K + Emotion \u2013 latency\/resolution matrix &amp; emotion propagation clarifications. (<a href=\"https:\/\/techcommunity.microsoft.com\/blog\/azure-ai-foundry-blog\/latest-avatar-updates-4k-support-and-emotion-control-features\/4449995\">avatar blog<\/a>)<\/li>\n<li>DeepSeek R1 \/ V3.1 \u2013 reasoning prompt engineering, deployment capability rows. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/foundry-models\/tutorials\/get-started-deepseek-r1\">DeepSeek tutorial<\/a>)<\/li>\n<li>Black Forest Labs &amp; FLUX.1 Kontext entries \u2013 parameter details (seed, aspect ratio) + safety notes. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/foundry-models\/includes\/models-azure-direct-others\">models include<\/a>)<\/li>\n<li>Retrieval service docs \u2013 knowledge sources abstraction &amp; answer synthesis latency benchmarks. (<a href=\"https:\/\/techcommunity.microsoft.com\/blog\/azure-ai-foundry-blog\/agentic-retrieval-updates-in-azure-ai-search\/4450621\">retrieval blog<\/a>)<\/li>\n<li>Evaluation service &amp; azure-ai-evaluation 1.11.0 \u2013 experiment tags, groundedness-with-tool-context, multilingual categories. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/concepts\/evaluation-evaluators\">evaluators<\/a>)<\/li>\n<li>Unified OpenAI client migration (Python\/JS\/.NET\/Java\/Go) \u2013 removal of legacy chat completions quickstarts; Responses API emphasized. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/openai\/includes\/language-overview\/python\">Python overview<\/a>)<\/li>\n<li>Function calling &amp; structured outputs \u2013 added gpt-5-codex; deprecated older GPT-4 preview variants. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/openai\/how-to\/function-calling\">function calling<\/a>)<\/li>\n<li>Content filtering configurability \u2013 threshold ranges &amp; alignment with safety annotations. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/openai\/how-to\/content-filters\">content filters<\/a>)<\/li>\n<li>Prompt engineering \u2013 warnings against extracting raw reasoning (policy risk) &amp; simplified scenario framing. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/openai\/concepts\/prompt-engineering\">prompt engineering<\/a>)<\/li>\n<li>Realtime API &amp; vision\/PDF guidance \u2013 generalized context length + image\/PDF capability scoping. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/openai\/how-to\/responses\">Responses API<\/a>)<\/li>\n<li>Fine-tuning overview &amp; model includes \u2013 GPT\u20114o\/mini availability; global training graduated from preview. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/concepts\/fine-tuning-overview\">fine-tuning overview<\/a>)<\/li>\n<li>RFT docs \u2013 custom graders security (AST evaluation), resource limits, removed deprecated fields. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/openai\/how-to\/reinforcement-fine-tuning\">RFT how-to<\/a>)<\/li>\n<li>Cost management (models\/platform\/hub) \u2013 unified, expanded budgeting &amp; export instructions; redirects added. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/foundry-models\/how-to\/manage-costs\">models cost mgmt<\/a>)<\/li>\n<li>Security &amp; governance TOC \u2013 reorganized hub vs project encryption, connections, quotas. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/concepts\/encryption-keys-portal\">encryption keys<\/a>)<\/li>\n<li>RBAC &amp; Entra ID guides \u2013 group management, subscription-scope purge visibility requirement. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/concepts\/rbac-azure-ai-foundry\">RBAC<\/a>)<\/li>\n<li>Network security perimeter \u2013 traffic routing order (Private Link precedence) &amp; diagnostics. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/how-to\/add-foundry-to-network-security-perimeter\">perimeter guide<\/a>)<\/li>\n<li>Deployment guides \u2013 portal\/Bicep\/CLI rewrites, endpoint retrieval clarity, template recommendations for private networking. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/foundry-models\/how-to\/create-model-deployments\">deployment guide<\/a>)<\/li>\n<li>Model router docs \u2013 removed limited access notes; clarified GPT-5 routing &amp; fallback logic. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/openai\/concepts\/model-router\">model router<\/a>)<\/li>\n<li>Model region &amp; quota matrices \u2013 added GPT-5, removed deprecated columns, standardized availability presentation. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/agents\/concepts\/model-region-support\">region support<\/a>)<\/li>\n<li>Quotas &amp; limits \u2013 standardized quota increase process include; added GPT-5 Codex, o-series, realtime entries. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/openai\/quotas-limits\">quotas &amp; limits<\/a>)<\/li>\n<li>Playgrounds &amp; first-run experiences \u2013 unified flows, expanded highlights, tracking parameters for analytics. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/concepts\/concept-playgrounds\">playgrounds<\/a>)<\/li>\n<li>Foundry Local architecture\/get-started \u2013 NPU (Intel\/AMD\/Qualcomm) support, driver notes, troubleshooting &amp; default model alias changes. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/foundry-local\/get-started\">Foundry Local get started<\/a>)<\/li>\n<li>Foundry Local CLI &amp; SDK refs \u2013 execution provider downloads, restart troubleshooting, EP options added. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/foundry-local\/reference\/reference-cli\">Foundry Local CLI<\/a>)<\/li>\n<li>Tracing &amp; evaluation guides \u2013 local collector workflow, playground trace\/eval clarity, evaluation results comparison instructions. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/how-to\/develop\/trace-application\">trace application<\/a>)<\/li>\n<li>Agents quickstarts &amp; concepts \u2013 removed redundant tables, expanded orchestration components, updated code tabs. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/agents\/quickstart\">agents quickstart<\/a>)<\/li>\n<li>Browser Automation &amp; Computer Use \u2013 troubleshooting, isolation model, risk categorization improvements. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/agents\/how-to\/tools\/browser-automation\">browser automation<\/a>)<\/li>\n<li>Bing tools &amp; samples \u2013 transaction counting, variable naming, configuration name terminology. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/agents\/how-to\/tools\/bing-custom-search\">bing tool<\/a>)<\/li>\n<li>Marketplace configuration \u2013 architecture diagrams &amp; troubleshooting expansions. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/foundry-models\/how-to\/configure-marketplace\">marketplace config<\/a>)<\/li>\n<li>Upgrade (Azure OpenAI \u2192 Foundry) \u2013 networking, pricing, troubleshooting, limitations clarifications; removed obsolete limitations. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/how-to\/upgrade-azure-openai\">upgrade guide<\/a>)<\/li>\n<li>Responses API, reasoning, structured outputs \u2013 \/openai\/v1 adoption, <code>model<\/code> payload field emphasis, unsupported feature pruning. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/openai\/how-to\/responses\">Responses API<\/a>)<\/li>\n<li>Model retirements timeline \u2013 new dates (o1-mini, Mistral variants) &amp; replacements (gpt-5-mini, updated large models). (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/openai\/concepts\/model-retirements\">model retirements<\/a>)<\/li>\n<li>Voice Live integration \/ gpt-realtime pairing \u2013 session state &amp; interruption handling examples. (<a href=\"https:\/\/techcommunity.microsoft.com\/blog\/azure-ai-foundry-blog\/upgrade-your-voice-agent-with-azure-ai-voice-live-api\/4458247\">Voice Live GA blog<\/a>)<\/li>\n<li>Agents continuous evaluation \u2013 project configuration steps &amp; region table. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/how-to\/continuous-evaluation-agents\">continuous evaluation<\/a>)<\/li>\n<li>FAQ &amp; legacy cleanups \u2013 removed outdated GPT\u20114 Turbo Q&amp;A; chat completions quickstarts deprecated &amp; redirected. (<a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-foundry\/openai\/faq\">FAQ<\/a>)<\/li>\n<\/ul>\n<h2>Join the community<\/h2>\n<p>Ship something with these updates and tell us about it! Join the discussion on <a href=\"https:\/\/aka.ms\/azureaifoundry\/discord\">Discord<\/a>, open a thread in <a href=\"https:\/\/aka.ms\/azureaifoundry\/forum\">GitHub Discussions<\/a> forum, or catch the next live deep\u2011dive (<a href=\"http:\/\/aka.ms\/ModelMondays\">Model Mondays<\/a> + Agent AMA).<\/p>\n","protected":false},"excerpt":{"rendered":"<p>GPT\u20115\u2011Codex GA, Sora video\u2011to\u2011video preview, Grok 4 Fast, Browser Automation &#038; Key Vault previews, new Search knowledge sources, Voice Live GA.<\/p>\n","protected":false},"author":185793,"featured_media":1307,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[1,27],"tags":[10,13,56,66,64,68,65,67,2,57,69],"class_list":["post-1305","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-microsoft-foundry","category-whats-new","tag-ai-agents","tag-azure-ai-search","tag-browser-automation","tag-evaluations","tag-gpt-5-codex","tag-identity-sdk","tag-key-vault","tag-knowledge-sources","tag-microsoft-foundry","tag-sora","tag-voicelive"],"acf":[],"blog_post_summary":"<p>GPT\u20115\u2011Codex GA, Sora video\u2011to\u2011video preview, Grok 4 Fast, Browser Automation &#038; Key Vault previews, new Search knowledge sources, Voice Live GA.<\/p>\n","_links":{"self":[{"href":"https:\/\/devblogs.microsoft.com\/foundry\/wp-json\/wp\/v2\/posts\/1305","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/devblogs.microsoft.com\/foundry\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/devblogs.microsoft.com\/foundry\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/foundry\/wp-json\/wp\/v2\/users\/185793"}],"replies":[{"embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/foundry\/wp-json\/wp\/v2\/comments?post=1305"}],"version-history":[{"count":0,"href":"https:\/\/devblogs.microsoft.com\/foundry\/wp-json\/wp\/v2\/posts\/1305\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/foundry\/wp-json\/wp\/v2\/media\/1307"}],"wp:attachment":[{"href":"https:\/\/devblogs.microsoft.com\/foundry\/wp-json\/wp\/v2\/media?parent=1305"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/foundry\/wp-json\/wp\/v2\/categories?post=1305"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/foundry\/wp-json\/wp\/v2\/tags?post=1305"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}