Microsoft’s AI history is difficult to follow because its major products were developed on different schedules and several changed names along the way.
Project Oxford became Cognitive Services, Bing Chat became Microsoft Copilot, and Azure AI Studio later developed into Microsoft Foundry. Microsoft also introduced its own Phi and MAI model families while expanding Copilot across Windows and Microsoft 365.
This timeline tracks the major milestones behind those changes from 2015 through 2026.
It records launch dates, model-family releases, public availability, renames, platform transitions, and major Microsoft–OpenAI partnership changes in chronological order.
Last updated: September 27, 2026.
Microsoft AI Timeline
2026
| Date | Milestone |
|---|---|
| Sep 25 | Microsoft introduced a new Copilot experience centered on Home, Code, and Autopilot. |
| Sep 4 | MAI-Image-2.6 and MAI-Image-2.6-Flash entered public preview in Microsoft Foundry with multi-image editing and web grounding. |
| Sep 3 | Microsoft introduced MAI-Transcribe-2 with transcription across 60 languages, speaker diarization, and word-level timestamps. |
| Jun 16 | Copilot Cowork reached general availability for Microsoft 365 Copilot customers worldwide. |
| Jun 2 | Microsoft AI expanded the MAI family with seven in-house models across reasoning, coding, image generation, voice, and transcription. |
| May 1 | Microsoft Agent 365 reached general availability as a control plane for observing, governing, and securing AI agents. |
| Apr 27 | Microsoft and OpenAI amended their agreement, making Microsoft’s model and product IP license nonexclusive and permitting OpenAI products across other cloud providers. |
| Apr 3 | Microsoft Agent Framework 1.0 reached production release for .NET and Python, unifying work from Semantic Kernel and AutoGen in one open-source agent SDK. |
| Apr 2 | MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2 became available to developers through Microsoft Foundry. |
| Mar 9 | Microsoft announced Wave 3 of Microsoft 365 Copilot, expanded model choice, and the May launch plan for Agent 365. |
| Jan 26 | Microsoft introduced Maia 200, its second-generation in-house AI accelerator, and began deploying it for Azure inference workloads. |
2025
| Date | Milestone |
|---|---|
| Nov 18 | Microsoft shifted Azure AI Foundry to the Microsoft Foundry brand as its platform for models, apps, and agents. |
| Nov 18 | Microsoft, NVIDIA, and Anthropic announced a partnership that expanded Claude availability through Azure and Microsoft Foundry. |
| Oct 28 | Microsoft and OpenAI signed a new definitive agreement extending Microsoft’s model and product IP rights through 2032 and changing future compute terms. |
| Aug 28 | Microsoft AI introduced MAI-Voice-1 and opened public testing for MAI-1-preview, its first foundation model trained end to end in-house. |
| May 19 | Azure AI Foundry Agent Service reached general availability for building and deploying managed AI agents. |
| Apr 23 | Microsoft expanded Microsoft 365 Copilot around agents, including Researcher, Analyst, the Agent Store, and Copilot Notebooks. |
| Apr 4 | Copilot Search in Bing began rolling out with AI-generated answers, conventional search results, and source links. |
| Jan 21 | Microsoft and OpenAI retained Azure API exclusivity while changing future training-compute capacity to a right-of-first-refusal model for Microsoft. |
| Jan 15 | Microsoft introduced Microsoft 365 Copilot Chat for commercial customers and added metered agents for eligible organizations. |
2024
| Date | Milestone |
|---|---|
| Dec 12 | Microsoft introduced Phi-4, a 14-billion-parameter small language model focused on complex reasoning. |
| Nov 19 | Microsoft introduced Azure AI Foundry as the next stage of Azure AI Studio and announced Azure AI Agent Service in preview. |
| May 21 | Azure AI Studio reached general availability for building, evaluating, deploying, and monitoring generative AI applications. |
| May 20 | Microsoft introduced Copilot+ PCs, a Windows PC category built around neural processing units for local AI workloads. |
| Apr 23 | Microsoft introduced the Phi-3 family, with Phi-3-mini available that day through Azure AI Studio, Hugging Face, and Ollama. |
| Jan 15 | Microsoft launched Copilot Pro and removed the 300-seat minimum for Microsoft 365 Copilot commercial subscriptions. |
2023
| Date | Milestone |
|---|---|
| Dec 12 | Microsoft released Phi-2 and placed the 2.7-billion-parameter model in the Azure AI Studio catalog. |
| Nov 15 | Bing Chat and Bing Chat Enterprise became Microsoft Copilot, with general availability following on December 1. |
| Nov 15 | Microsoft introduced Copilot Studio as the successor to Power Virtual Agents for building and customizing business copilots and agents. |
| Nov 15 | Azure AI Studio entered public preview for generative AI application development. |
| Nov 15 | Microsoft announced Maia 100, its first in-house AI accelerator for Azure datacenters. |
| Nov 1 | Microsoft 365 Copilot reached general availability for enterprise customers. |
| Sep 21 | Microsoft introduced a unified Copilot identity across Windows, Bing, Edge, and Microsoft 365, with the first Windows rollout beginning September 26. |
| Jun | Microsoft Research introduced Phi-1, a 1.3-billion-parameter code model that began the Phi model family. |
| Mar 16 | Microsoft introduced Microsoft 365 Copilot for Word, Excel, PowerPoint, Outlook, Teams, and other work apps. |
| Feb 7 | Microsoft launched the new AI-powered Bing and Edge in limited preview with conversational search and chat. |
| Jan 23 | Microsoft and OpenAI announced a multiyear, multibillion-dollar expansion of their partnership covering Azure supercomputing, research, and product deployment. |
| Jan 17 | Azure OpenAI Service reached general availability. |
2019–2021
| Date | Milestone |
|---|---|
| Nov 2, 2021 | Microsoft announced Azure OpenAI Service with invitation-only access. |
| Sep 22, 2020 | Microsoft obtained an exclusive license to use GPT-3 in its own products and services while OpenAI continued operating its API. |
| Jul 22, 2019 | Microsoft invested $1 billion in OpenAI and expanded the partnership around Azure AI supercomputing and commercial deployment. |
2015–2016
| Date | Milestone |
|---|---|
| Nov 15, 2016 | OpenAI selected Microsoft Azure as its primary cloud platform for large-scale deep-learning and AI research. |
| Mar 30, 2016 | Microsoft introduced Cognitive Services, expanding Project Oxford into a larger set of vision, speech, language, knowledge, and search APIs. |
| Apr 30, 2015 | Microsoft introduced Project Oxford in beta with APIs and SDKs for adding AI capabilities to applications. |
Major Microsoft AI Milestones
Bing Chat becomes Microsoft Copilot
The February 2023 Bing launch placed conversational AI directly inside Microsoft’s search product. Microsoft expanded the Copilot identity across Windows, Bing, Edge, and Microsoft 365 later that year, followed by the formal Bing Chat rename in November.
Copilot developed into a product identity beyond Bing. Bing retained its own search experience and later introduced Copilot Search, while the standalone Copilot product expanded across consumer and workplace use.
Microsoft 365 Copilot develops into an agent platform
Microsoft 365 Copilot launched around AI assistance inside Word, Excel, PowerPoint, Outlook, Teams, and Microsoft Graph data. Copilot Studio extended the ecosystem to custom business agents.
The product expanded further in 2025 and 2026 through Researcher, Analyst, Cowork, Agent 365, and a wider set of model choices. Microsoft 365 Copilot now includes both interactive assistance and longer-running delegated work.
Copilot+ PCs establish a Windows AI hardware category
Microsoft introduced Copilot+ PCs in May 2024 around Windows devices equipped with neural processing units for local AI workloads.
The category created a hardware boundary for Windows AI features that depend on local inference. Copilot itself remains available across a wider set of Windows systems, while selected on-device AI experiences require compatible Copilot+ hardware.
Phi grows from a research model into Microsoft’s small-model family
Phi-1 began in Microsoft Research in June 2023 as a compact code model trained with curated and synthetic data. Phi-2 moved the lineage into Azure AI Studio later that year.
Phi-3 expanded the family into models distributed through Azure, Hugging Face, and Ollama, while Phi-4 pushed the line further into reasoning. Later Phi releases extended the family into multimodal and reasoning-focused variants.
Azure AI Studio evolves into Microsoft Foundry
Azure AI Studio entered public preview in November 2023 and reached general availability in May 2024. Azure AI Foundry followed later that year with a wider platform centered on models, application development, evaluation, and agents.
The Microsoft Foundry name arrived in 2025 as the platform expanded beyond the earlier Azure AI Studio identity. Foundry Agent Service handles managed agents, while Microsoft Agent Framework provides an open-source SDK and runtime for agent development.
MAI becomes Microsoft’s first-party model portfolio
Microsoft AI introduced MAI-Voice-1 and MAI-1-preview in August 2025. The model line expanded into speech, image generation, reasoning, and coding over the following year.
Microsoft Foundry became a developer distribution point for MAI models in 2026. MAI-Transcribe-2 and MAI-Image-2.6 continued the model lineage in September with new speech and image-generation releases.
The Microsoft and OpenAI partnership changes over time
OpenAI began using Azure as its primary cloud platform for large-scale AI research in 2016. Microsoft’s 2019 investment expanded the relationship into Azure supercomputing and commercial deployment, followed by Azure OpenAI Service and deeper product integration.
The commercial terms changed substantially in 2025 and 2026. Microsoft’s model and product IP rights were extended through 2032, its right of first refusal over new OpenAI compute capacity was removed, and the April 2026 amendment made Microsoft’s license nonexclusive. Azure remains OpenAI’s primary cloud partner, while OpenAI can serve its products through other cloud providers.
Related Resources
- OpenAI & ChatGPT Timeline: GPT Release Dates to GPT-6 Astra (2026)
- AI Model Release Calendar: Latest AI Model Releases by Date
- GitHub Copilot Cheat Sheet: CLI & Slash Commands, Variables, and Shortcuts
