this article summarizes recent releases and documentation updates for Azure OpenAI in Azure artificial intelligence Foundry models. the Sora model now supports video-to-video generation and image-to-video generation, allowing users to provide a short video or a single image as input and generate longer videos that incorporate that content. Sora availability expanded to the Sweden Central and East US 2 regions. the article points readers to the video-generation quickstart for implementation details.
OpenAI´s realtime and audio models are now generally available on Azure AI Foundry direct models. notable model improvements include better instruction following for tone, pacing, and escalation, multilingual switching, new standard voices named Marin and Cedar, and improved audio quality with glitch-free output and better alphanumeric reproduction. realtime API service upgrades include support for image input via voice, improved function calling with async support, a conversation mode that uses voice activity detection for natural turn-taking, and SIP entry with public SIP URI, SIPS/SRTP, PSTN, and HD Voice. the documentation recommends transitioning to the GA models and exploring capabilities in the Azure OpenAI documentation and Azure AI Foundry Playground.
the release stream also covers provisioning and model management changes. provisioned spillover reached general availability to route traffic overages from provisioned deployments to standard deployments. the GPT-5 series is available in several sizes—gpt-5, gpt-5-mini, gpt-5-nano, and gpt-5-chat—with registration required for access to the full gpt-5 model while the smaller variants and chat version do not require registration. a new version of model-router now supports the GPT-5 series but is limited access; interested users can request access via the gpt-5 access form. the update log also highlights prior 2025 releases including GPT-image-1 preview, codex-mini and o3-pro models, o3-mini, GPT 4.1 and multiple GPT-4o audio models, realtime WebRTC preview support, stored completions, and preference fine-tuning previews.
