Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Applies only to: Foundry (classic) portal . This article isn't available for the new Foundry portal. Learn more about the new portal . Links in this article might open content in the new Microsoft Foundry documentation instead of the Foundry (classic) documentation you're viewing now. This article provides a summary of the latest releases and major documentation updates for Azure OpenAI. GPT Realtime 2.0 concept articles New concept articles describe the capabilities and key concepts for the latest realtime model: GPT Realtime 2.0 (preview) overview — covers reasoning support, response phases (preamble and final answer), and stricter instruction following. GPT Realtime Translate and GPT Realtime Whisper New concept articles describe the capabilities and use cases for the latest real-time translation and transcription models: GPT Realtime Translate — covers continuous stream-based audio translation for live multilingual events, customer support, and voice assistants. GPT Realtime Whisper — covers low-latency streaming transcription for live captions, monitoring, and archival workflows. GPT-Realtime-1.5 and GPT-Audio-1.5 models released The gpt-realtime-1.5 and gpt-audio-1.5 models are now available. These models build on last year's GPT-Realtime and GPT-Audio with improvements in instruction following, multilingual support, and tool calling, while preserving the low-latency, real-time interactions developers need for voice-based applications. Try them with the existing chat completion APIs in Microsoft Foundry. GPT-image-1.5 model is now available GPT-image-1.5 is OpenAI's latest cutting-edge image generation model. It features improved performance, quality, editing controls, and face preservation. In editing mode, the model supports high input_fidelity and adding or removing one aspect of the input image while retaining others. Request access: limited access model application Key model capabilities: Includes all capabilities of GPT-image-1: Text to image generation Image to image generation (editing) Text to image generation Image to image generation (editing) High quality image generations, up to 1024x1536 and 1536x1024 pixels Face preservation Follow the image generation how-to guide to get started with this model. Automatic speech recognition (ASR) model update gpt-4o-mini-transcribe-2025-12-15 Improved transcription accuracy and robustness for real-time scenarios, with approximately 50% lower word error rate (WER) than the previous gpt-4o-transcribe-mini model on English benchmarks. Improves multilingual performance across Japanese, Indic, and other languages. Reduces hallucinations on silence by up to 4×, making it a more reliable choice for noisy environments and real-world audio streams. Input remains audio, with text as output, and deployment is API-only. Realtime-mini (speech-to-speech) model update gpt-realtime-mini-2025-12-15 Feature parity with full gpt-realtime model in instruction-following and function-calling. Input and output are both audio, and deployment is API-only. Text to speech model update gpt-4o-mini-tts-2025-12-15 Sets a new benchmark for multilingual speech synthesis. Produces more natural, human-like speech with fewer artifacts and improved speaker similarity. Input is text, output is audio, and deployment is API-only. Realtime API support for SIP The Realtime API now supports SIP, enabling telephony connections to realtimeapi. For more information, see the Realtime SIP documentation . GPT-4o audio model released The gpt-4o-transcribe-diarize speech to text model is released. This model is an Automatic Speech Recognition (ASR) model that converts spoken language into text in real time. It enables organizations to unlock insights from conversations instantly with ultra-low latency and high accuracy across 100+ languages. This capability is essential for workflows where voice data drives decisions—such as customer support, virtual meetings, and live events. Diarization is the process of identifying who spoke when in an audio stream. It transforms conversations into speaker-attributed transcripts, enabling businesses to extract actionable insights from meetings, customer calls, and live events. With advanced models like gpt-4o-transcribe-diarize , organizations gain real-time clarity and context—turning voice into structured data that drives smarter decisions and improves productivity, supporting automatic speech recognition. Use this model via the /audio and /realtime APIs. GPT-image-1-mini The gpt-image-1-mini model is now available for global deployments. It's a smaller version of the gpt-image-1 model that offers a good balance between performance and cost. All use cases are currently supported, except for image edits and input fidelity. Request access: Limited access model application Follow the image generation how-to guide to get started with this model. PII detection content filter Personally identifiable information (PII) detection is now available as a built-in content filter. This feature allows you to identify and block sensitive information in LLM outputs, enhancing data privacy. For more information, see the PII detection documentation. GPT-5-codex is now available To learn more about gpt-5-codex , see the getting started with reasoning models page . To learn more about gpt-5-codex , see the getting started with reasoning models page . gpt-5-codex is designed for use with the Codex CLI and the Visual Studio Code Codex extension . gpt-5-codex is designed for use with the Codex CLI and the Visual Studio Code Codex extension . Registration is required for access to the gpt-5-codex model . If you have previously registered and obtained access to other limited access models like gpt-5 , you don't need to reapply and will automatically be granted access. Registration is required for access to the gpt-5-codex model . If you have previously registered and obtained access to other limited access models like gpt-5 , you don't need to reapply and will automatically be granted access. Sora video-to-video support The Sora model from OpenAI now supports video-to-video generation. You can provide a short video as input to generate a new, longer video that incorporates the input video. See the quickstart to get started. Sora image-to-video support The Sora model from OpenAI now supports image-to-video generation. You can provide an image as input to the model to generate a video that incorporates the content of the image. You can also specify the frame of the video in which the image should appear: it doesn't need to be the beginning. See the quickstart to get started. This Sora model is now available in the Sweden Central region and East US 2. Realtime API audio model GA OpenAI's GPT RealTime and Audio models are now generally available in Microsoft Foundry Models. Model improvements: Improved instruction following: Enhanced capabilities to follow tone, pacing, and escalation instructions more accurately and reliably. Can also switch languages. New standard voices, Marin and Cedar, that bring improved naturalness and clarity to speech synthesis. Improved audio quality: Glitch-free output, improved alphanumeric reproduction, and modality control. Realtime API service improvements: Image Input: Add and discuss images via voice without video requirements. Improved function calling: Enhanced ability to call custom code defined by developers. Async function calling is supported, allowing sessions to continue while a function call is pending. Conversation Mode: Real-world turn-taking behavior for natural interactions. Conversation mode uses VAD to prompt users if no response is detected, improving real-world usability for phone-like interactions. We highly recommend that all customers transition to the newly launched GA models to take full advantage of the latest features. Visit the Azure OpenAI documentation and Microsoft Foundry portal to explore capabilities and integrate into your applications. Provisioned spillover General Availability (GA) Spillover is now Generally Available. Spillover manages traffic fluctuations on provisioned deployments by routing overages to a designated standard deployment. To learn more about how to maximize utilization for your provisioned deployments with spillover, see Manage traffic with spillover for provisioned deployments . GPT-5 models available gpt-5 , gpt-5-mini , gpt-5-nano To learn more, see the getting started with reasoning models page . gpt-5 , gpt-5-mini , gpt-5-nano To learn more, see the getting started with reasoning models page . gpt-5-chat is now available. To learn more, see the models page gpt-5-chat is now available. To learn more, see the models page gpt-5 is now available for Provisioned Throughput Units (PTU) . gpt-5 is now available for Provisioned Throughput Units (PTU) . Registration is required for access to the gpt-5 model . Registration is required for access to the gpt-5 model . gpt-5-mini , gpt-5-nano , and gpt-5-chat don't require registration. gpt-5-mini , gpt-5-nano , and gpt-5-chat don't require registration. New version of model-router Model router now supports GPT-5 series models. Model router now supports GPT-5 series models. Model router for Microsoft Foundry is a deployable AI chat model that automatically selects the best underlying chat model to respond to a given prompt. For more information on how model router works and its advantages and limitations, see the Model router concepts guide . To use model router with the Completions API, follow the How-to guide . Model router for Microsoft Foundry is a deployable AI chat model that automatically selects the best underlying chat model to respond to a given prompt. For more information on how model router works and its advantages and limitations, see the Model router concepts guide . To use model router with the Completions API, follow the How-to guide .

GPT-image-1 update (preview) Input fidelity parameter: The input_fidelity parameter in the image edits API lets you control how closely the model conveys the style and features of the subjects in the original (input) image. This is useful for: Editing photos while preserving facial features; creating avatars that look like original person across different styles; combining faces from multiple people into one image. Maintaining brand identity in generated images for marketing assets, mockups, product photography. E-commerce and fashion, where you need to edit images of outfits or product details without compromising realism. Input fidelity parameter: The input_fidelity parameter in the image edits API lets you control how closely the model conveys the style and features of the subjects in the original (input) image. This is useful for: Editing photos while preserving facial features; creating avatars that look like original person across different styles; combining faces from multiple people into one image. Maintaining brand identity in generated images for marketing assets, mockups, product photography. E-commerce and fashion, where you need to edit images of outfits or product details without compromising realism. Partial image streaming: The image generation and image edits APIs support partial image streaming, where they return images with partially rendered content throughout the image generation process. Display these images to the user to provide earlier visual feedback and show the progress of the image generation operation. Partial image streaming: The image generation and image edits APIs support partial image streaming, where they return images with partially rendered content throughout the image generation process. Display these images to the user to provide earlier visual feedback and show the progress of the image generation operation. New models codex-mini and o3-pro released codex-mini and o3-pro are now available. To learn more, see the getting started with reasoning models page . Sora video generation released (preview) The Sora (2025-05-02) model is a video generation model from OpenAI that can create realistic and imaginative video scenes from text instructions. Follow the Video generation quickstart to get started. For more information, see the Video generation concepts guide. Spotlighting for prompt shields (preview) Spotlighting is a sub-feature of prompt shields that enhances protection against indirect (embedded document) attacks by tagging input documents with special formatting to indicate lower trust to the model. For more information, see the Prompt shields filter documentation. Model router (preview) Model router for Microsoft Foundry is a deployable AI chat model that automatically selects the best underlying chat model to respond to a given prompt. For more information on how model router works and its advantages and limitations, see the Model router concepts guide . To use model router with the Completions API, follow the How-to guide . Realtime API (preview) support for WebRTC The Realtime API (preview) now supports WebRTC, enabling real-time audio streaming and low-latency interactions. This feature is ideal for applications requiring immediate feedback, such as live customer support or interactive voice assistants. For more information, see the Realtime API (preview) documentation . GPT-image-1 released (preview, limited access) GPT-image-1 ( 2025-04-15 ) is the latest image generation model from Azure OpenAI. It features major improvements over DALL-E, including: Better at responding to precise instructions. Reliably renders text. Accepts images as input, which enables the new capabilities of image editing and inpainting. Request access: Limited access model application Follow the image generation how-to guide to get started with the new model. o4-mini and o3 models released o4-mini and o3 models are now available. These models are the latest reasoning models from Azure OpenAI offering enhanced reasoning, quality, and performance. For more information, see the getting started with reasoning models page . GPT-4.1 released GPT 4.1 and GPT 4.1-nano are now available. These models are the latest models from Azure OpenAI. GPT 4.1 has a 1 million token context limit. For more information, see the models page . gpt-4o audio models released New audio models powered by GPT-4o are now available. The gpt-4o-transcribe and gpt-4o-mini-transcribe speech to text models are released. Use these models via the /audio and /realtime APIs. The gpt-4o-transcribe and gpt-4o-mini-transcribe speech to text models are released. Use these models via the /audio and /realtime APIs. The gpt-4o-mini-tts text to speech model is released. Use the gpt-4o-mini-tts model for text to speech generation via the /audio API. The gpt-4o-mini-tts text to speech model is released. Use the gpt-4o-mini-tts model for text to speech generation via the /audio API. For more information about available models, see the models and versions documentation . Responses API and computer-use-preview model The Responses API is a stateful API from Azure OpenAI that provides a unified experience for model interactions. The Responses API also adds support for the new computer-use-preview model, which powers the Computer use capability. For access to computer-use-preview registration is required, and access will be granted based on Microsoft's eligibility criteria . Customers who have access to other limited access models still need to request access for this model. Request access: computer-use-preview limited access model application For more information on model capabilities, and region availability see the models documentation . Playwright integration demo code . Provisioned spillover (preview) Spillover manages traffic fluctuations on provisioned deployments by routing overages to a designated standard deployment. To learn more about how to maximize utilization for your provisioned deployments with spillover, see Manage traffic with spillover for provisioned deployments (preview) . Specify content filtering configurations In addition to the deployment-level content filtering configuration, we now also provide a request header that allows you specify your custom configuration at request time for every API call. For more information, see Use content filters (preview) . GPT-4.5 Preview The latest GPT model that excels at diverse text and image tasks is now available on Azure OpenAI. For more information on model capabilities, and region availability see the models documentation . Stored completions API Stored completions allow you to capture the conversation history from chat completions sessions to use as datasets for evaluations and fine-tuning. o3-mini data zone standard deployments o3-mini is now available for global standard, and data zone standard deployments for registered limited access customers. For more information, see our reasoning model guide . gpt-4o mini audio released The gpt-4o-mini-audio-preview ( 2024-12-17 ) model is the latest audio completions model. For more information, see the audio generation quickstart . The gpt-4o-mini-realtime-preview ( 2024-12-17 ) model is the latest real-time audio model. The real-time models use the same underlying GPT-4o audio model as the completions API, but is optimized for low-latency, real-time audio interactions. For more information, see the real-time audio quickstart . For more information about available models, see the models and versions documentation . o3-mini released o3-mini ( 2025-01-31 ) is the latest reasoning model, offering enhanced reasoning abilities. For more information, see our reasoning model guide . GPT-4o audio completions The gpt-4o-audio-preview model is now available for global deployments in East US 2 and Sweden Central regions . Use the gpt-4o-audio-preview model for audio generation. The gpt-4o-audio-preview model introduces the audio modality into the existing /chat/completions API. The audio model expands the potential for AI applications in text and voice-based interactions and audio analysis. Modalities supported in gpt-4o-audio-preview model include: text, audio, and text + audio. For more information, see the audio generation quickstart . The Realtime API uses the same underlying GPT-4o audio model as the completions API, but is optimized for low-latency, real-time audio interactions. GPT Realtime API 2024-12-17 The gpt-4o-realtime-preview model version 2024-12-17 is available for global deployments in East US 2 and Sweden Central regions . Use the gpt-4o-realtime-preview version 2024-12-17 model instead of the gpt-4o-realtime-preview version 2024-10-01-preview model for real-time audio interactions. Added support for prompt caching with the gpt-4o-realtime-preview model. Added support for new voices. The gpt-4o-realtime-preview models now support the following voices: alloy , ash , ballad , coral , echo , sage , shimmer , verse . Rate limits are no longer based on connections per minute. Rate limiting is now based on RPM (requests per minute) and TPM (tokens per minute) for the gpt-4o-realtime-preview model. The rate limits for each gpt-4o-realtime-preview model deployment are 100 K TPM and 1 K RPM. During the preview, Azure AI Foundry portal and APIs might inaccurately show different rate limits. Even if you try to set a different rate limit, the actual rate limit is 100 K TPM and 1 K RPM. For more information, see the GPT real-time audio quickstart and the how-to guide . o1 reasoning model released for limited access The latest o1 model is now available for API access and model deployment. Registration is required, and access will be granted based on Microsoft's eligibility criteria . Customers who previously applied and received access to o1-preview , don't need to reapply as they're automatically on the wait-list for the latest model. Request access: limited access model application To learn more about the advanced o1 series models see, getting started with o1 series reasoning models . Region availability Preference fine-tuning (preview) Direct preference optimization (DPO) is a new alignment technique for large language models, designed to adjust model weights based on human preferences. Unlike reinforcement learning from human feedback (RLHF), DPO doesn't require fitting a reward model and uses simpler data (binary preferences) for training. This method is computationally lighter and faster, making it equally effective at alignment while being more efficient. DPO is especially useful in scenarios where subjective elements like tone, style, or specific content preferences are important. We're excited to announce the public preview of DPO in Azure OpenAI, starting with the gpt-4o-2024-08-06 model. For fine-tuning model region availability, see the models page . Stored completions and distillation Stored completions allow you to capture the conversation history from chat completions sessions to use as datasets for evaluations and fine-tuning . GPT-4o 2024-11-20 gpt-4o-2024-11-20 is now available for global standard deployment in: North Central US South Central US NEW data zone provisioned deployment type Data zone provisioned deployments are available in the same Azure OpenAI resource as all other Azure OpenAI deployment types but allow you to use Azure global infrastructure to dynamically route traffic to the data center within the Microsoft defined data zone with the best availability for each request. Data zone provisioned deployments provide reserved model processing capacity for high and predictable throughput using Azure infrastructure within Microsoft specified data zones. Data zone provisioned deployments are supported on gpt-4o-2024-08-06 , gpt-4o-2024-05-13 , and gpt-4o-mini-2024-07-18 models. For more information, see the deployment types guide . Learn more about the underlying models that power Azure OpenAI . Was this page helpful? Need help with this topic? Want to try using Ask Learn to clarify or guide you through this topic? Additional resources Last updated on 2026-09-10