📋About Microsoft
Updated September 22, 2026Microsoft Corporation is one of the world's most valuable companies, founded by Bill Gates and Paul Allen in 1975. Windows, Microsoft 365, Azure, LinkedIn, GitHub and Xbox touch billions of users and virtually every enterprise on earth.
Microsoft made the largest corporate bet on AI in history through its partnership with OpenAI, investing over 13 billion dollars, and has integrated AI across every major product: Microsoft 365 Copilot in Word, Excel, PowerPoint, Outlook and Teams; GitHub Copilot, the most widely used AI coding tool; and Azure AI for enterprise infrastructure and model hosting. The strategy leverages a distribution advantage no competitor has — over 1.5 billion Office users, tens of millions of GitHub developers and the second-largest cloud platform — letting Microsoft deploy AI to enterprise users faster than anyone else, with Azure the default cloud for AI startups and enterprises alike.
Increasingly it also builds frontier AI in-house. The first-party MAI family spans reasoning, coding, image, voice, transcription and cybersecurity, including a compact security model that hunts vulnerabilities in large codebases and powers the company's vulnerability-remediation system, absorbing most tasks and routing only the hardest to a larger frontier model. Owning the models rather than only distributing a partner's gives Microsoft control over cost, availability and product integration, and hedges long-term dependence on any single lab even as the OpenAI partnership remains central. The same ambition extends to agent-first device platforms and to topological quantum computing, on a roadmap toward a commercially useful machine by 2029.
That multi-model posture is visible in shipping products. The agentic layer inside Microsoft 365 Copilot plans and executes long-running, multi-tool tasks and deliberately routes across Anthropic's and OpenAI's models, with usage-based credits billed on top of the license; Microsoft has disclosed it is evaluating self-hosted open-weight models on Azure as a lower-cost engine for cost-sensitive workloads. The driver is economics — agentic AI makes hundreds of model calls per task, so cheaper options hedge exposure to frontier inference costs. Microsoft has also stood up a dedicated operating business that embeds its own engineers inside customer operations to design and run their AI systems, mirroring forward-deployed units at Amazon, OpenAI and Anthropic and reflecting that the hardest part of enterprise AI is the last mile rather than the model.
The strategic direction is consistent: own more of the stack rather than only distribute a partner's technology. Microsoft has also published a draft Humanist AI Code of Conduct governing how its in-house models should behave, placing absolute constraints and human-control requirements above anything a customer can configure, and forbidding a model from evading oversight. It takes a position most labs avoid stating outright — that AI is not conscious, should not be designed to imitate consciousness, and should not be pursued as a legal person or treated as deserving welfare — a direct break with the model-welfare research Anthropic pursues.
🛠️Products & Tools (21)
Reading-accessibility tool built into Microsoft apps — text-to-speech, line focus, syllable and part-of-speech breakdown, spacing, and translation to support readers of all abilities.
Microsoft's unified enterprise platform for building and governing AI agents. Replaces AutoGen (2026). Consolidates agent development under a single framework with enterprise security and compliance.
Microsoft's chip-to-cloud platform for agent-first devices, unveiled at Build 2026 by Steven Bathiche. Builds hardware where AI agents — not apps — are the primary interface, on an enterprise Android base (MDEP) with reference designs. In early pilots with CVS Health, Target, Best Buy, Levi's, and AccuWeather.
AI pair programmer across VS Code, JetBrains and other IDEs — code completions, chat, multi-file edits, agent mode and an async coding agent that works on GitHub Issues. Offers model choice across four providers, plus runtime multi-model orchestration via Project HydraFusion.
MAI-Code-1-Flash is Microsoft's lightweight, in-house coding model built end-to-end for GitHub Copilot. Microsoft says it outperforms Anthropic's Claude Haiku 4.5 across its coding benchmarks — including a roughly 16-point lead on SWE-Bench Pro (51 percent versus 35 percent) — while using up to 60 percent fewer tokens. An adaptive-thinking mechanism scales reasoning effort to task complexity, and it rolls out to Visual Studio Code Copilot users via the model picker.
Copilot built into Microsoft Edge. Summarize pages, compare products, generate text in sidebars, and ask questions about the current page content.
Microsoft's cloud storage integrated with Windows 11 and Microsoft 365. Copilot AI can analyze files stored in OneDrive via Word, Excel, and Teams.
Microsoft Copilot's screen-reading capability that can see and understand what you're looking at on screen and help complete tasks in real time.
Microsoft's small-but-powerful open model family (MIT license). Phi-4 excels at reasoning and math at small size. Variants: Phi-4-multimodal (5.6B, speech+vision+text), Phi-4-mini (3.8B), Phi-4-reasoning (14B). On-device capable.
MAI-Thinking-1 is Microsoft's first in-house frontier reasoning model, unveiled at Build 2026. A roughly 35-billion-parameter model with a 256K-token context window, Microsoft says it matches Anthropic's Claude Opus 4.6 on the SWE-Bench Pro coding benchmark. It is the flagship of a new family of seven first-party MAI models and is in private preview on Azure AI Foundry.
Microsoft's strongest in-house image generation + editing model (Build 2026). Microsoft says it launched No. 2 for image editing on LMArena, with standout text rendering + identity consistency. Live in PowerPoint/Copilot + Microsoft Foundry; a faster Flash variant handles high-volume production.
Generative-AI assistant for security teams — investigates and summarizes incidents, triages alerts, and recommends responses in natural language across the Microsoft security stack, with autonomous agents.
Microsoft's first in-house cybersecurity model — a compact, code-heavy model that finds vulnerabilities in large codebases and powers the MDASH remediation system, reported at 96 percent on CyberGym at half the cost of the frontier stack it replaces.
Microsoft MatterGen is a generative-AI diffusion model that designs novel inorganic materials to order — prompt it with target properties and it proposes new, stable crystal structures; published in Nature and released open-source, paired with the MatterSim property predictor.
Microsoft clinical AI assistant (formerly Nuance DAX) that listens to patient-physician encounters and auto-generates SOAP notes, visit summaries, referral letters, and after-visit instructions. Agentic capabilities handle prior authorizations and medication refills. Deployed at 600+ health systems.
AI assistant embedded in Word, Excel, PowerPoint, Outlook, and Teams. Draft documents, analyze spreadsheets, generate presentations, and summarize meetings.
Microsoft's agent for delegated, multi-step work inside Microsoft 365 Copilot's Home tab, routing across Anthropic's Claude and OpenAI's GPT models, grounded in org data through Work IQ and billed by usage in Copilot Credits.
Microsoft's second-generation topological quantum chip, unveiled at Build 2026 with a reported ~1,000-fold qubit-reliability gain over Majorana 1. Notably designed and fabricated with help from Microsoft's agentic Discovery platform — an example of AI accelerating frontier science. The underlying physics remains contested.
Microsoft's AI assistant, built into Windows, Edge and Microsoft 365 and rebuilt in September 2026 around Home, Code and Autopilot tabs, drawing on models from OpenAI and Anthropic.
Microsoft's in-house text-to-speech model, released October 1, 2026 as the successor to MAI-Voice-2. Speaks 23 languages across 26 locales, lets one voice keep a native accent in each, and clones a voice from a few seconds of audio behind consent checks. $22 per million characters, or $15 for the Flash version built for live voice agents. In Microsoft Foundry.
Microsoft's in-house speech-to-text model, released September 3, 2026. Microsoft reports 60 languages at a 5.2% average word error rate on FLEURS, and 10x the speed of GPT-Transcribe, 7x ElevenLabs Scribe v2 and 5x Gemini 3.5 Transcribe. 10 cents/hour as a limited-time rate through end of 2026. A real-time Streaming version followed on October 1, 2026, returning first words in about 100 milliseconds at 54 cents/hour. Note the 5.2% is not comparable to v1.5's 2.4%, which averaged over 43 languages rather than 60.
Keep track of the companies you’re watching
- Save the companies you want to follow
- Get ⚡ alerts when your saved companies change
- Every product they ship, cross-linked to 900+ AI tool profiles
- Today’s top AI Stories — the day’s most important AI news, free
Swipe for Recommended for you and My AI Tools
Your AI Hub — sample data. See desktop view example