Timeline

Aion Instruct is planned to roll out to retail Windows devices, completing the transition from Phi Silica.

Consumption-based billing begins for Copilot harness runtime, including sandbox compute and memory.

Aion Instruct is planned to begin rolling out to Windows Insider Preview devices.

BioNeMo skills power a protein binder-design agent available through the Microsoft Discovery community catalog.

Copilot in Database Hub becomes available in public preview as an opt-in experience.

Azure Content Understanding adds support for three model families across document, text, speech, and video workflows.

.NET Conf runs November 10–13, 2026.

Multilingual post-stream refinement becomes generally available with a native LLM Speech streaming model.

Kimi K3.1 Pro becomes available through Fireworks on Microsoft Foundry.

Agentic mode is available in preview for multi-step document extraction using document-specific tools.

Real-Time Multichannel Speech-to-Text becomes generally available with per-channel language identification.

The Durable Task extension becomes generally available for durable agent sessions and workflows.

Claude Fable 5.2 becomes available in Microsoft Foundry as a generally available model.

Claude Fable 5.2 Fast becomes available in Microsoft Foundry as a generally available model.

Claude Mythos 5.2 is introduced in Microsoft Foundry with access restricted to eligible customers.

Support ends for the earlier Fabric data agent custom-tool integration; affected agents must migrate to Fabric IQ.

Microsoft Foundry adds hosted fal video models, including MiniMax H3 Max.

Kling, LTX, and MiniMax video models begin rolling out in public preview as a first-party Azure service.

Copilot harness enters public preview with a managed agent loop and sandbox.

Moonshot AI’s Kimi K3.1 Pro launches in Microsoft Foundry with text and image understanding and up to one million tokens of context.

Kling, LTX, and MiniMax video models begin rolling out in public preview through fal as a first-party Azure service.

Installable builder skills provide coding agents with Microsoft-maintained Agent Framework implementation guidance for .NET and Python.

Advanced DAX generation begins rolling out to the standard runtime, with completion expected November 10.

Advanced NL2SQL begins rolling out to the standard runtime, with completion expected November 10.

Voice Live API gains SIP support for direct telephony connections.

Voice agents gain direct SIP connectivity for calls through existing telephony infrastructure.

The Microsoft Copilot harness becomes available in public preview for managed agent execution.

Long-running resilience for hosted agents becomes generally available.

MAI-Image-2.6 becomes generally available in Microsoft Foundry.

MAI-Image-2.6-Flash becomes generally available in Microsoft Foundry.

LLM Speech 09-2026 is released through the Fast Transcription API with 25 additional languages and improved recognition and processing speed.

Agent lifecycle management for Copilot Chat agents reaches general availability, supporting separate versions across environments.

Connected agents for Copilot Chat agents reach general availability, enabling delegation to other agents.

Evaluations for Copilot Chat agents reach general availability for structured agent quality assessment.

Topics for Copilot Chat agents reach general availability, supporting deterministic conversational flows.

GPT-6.1 Terra becomes generally available in Microsoft Foundry for visual reasoning and computer-use workloads.

Multi-LoRA adapter deployment enters preview, allowing compatible fine-tuned adapters to share a base-model deployment.

The Decisions API becomes generally available in Microsoft Foundry for structured decisions within AI applications and agentic workflows.

Flint-interactive enables AI agents to generate interactive data visualizations using semantic events and update operators.

Managed Fine-Tuning adds four open-source models and MAI-Code-1.1-Flash in public preview.

Shared Multi-LoRA deployments enter public preview, pooling compatible adapters on common base-model weights and GPU capacity.

Cohere Rerank V5 becomes available in Microsoft Foundry with Fast and Pro variants and Context Focus.

Browser Automation Tool becomes generally available with managed cloud browsers for agent-driven web workflows.

Microsoft deprecates the earlier custom-connection integration for Fabric data agents.

Automation flow evaluations became generally available, extending structured quality assessment to automation flows.

The tool accuracy grader becomes generally available for evaluating agent tool selection and input accuracy.

Experimental llama.cpp support becomes available for local GGUF inference through task-specific APIs.

The experimental Windows-native Runtime API enters preview for controlling model execution and composition.

A standalone Aion Instruct package becomes available for developers to test applications and customizations.

Azure DocumentDB MCP Toolkit enters public preview, providing AI agents with MCP access to Azure DocumentDB.