Microsoft's 7 MAI Models at Build 2026

Why Microsoft built its own models: $13B dependency and contract freedom in late 2025

Over seven years, Microsoft invested more than $13 billion in OpenAI. GPT models on Azure became the backbone of its AI strategy. That deep dependency created three structural risks:

1. 01

   Runaway costs: Every API call pays OpenAI. Scale up and margins shrink.

2. 02

   No technical sovereignty: Microsoft cannot control model iteration pace, data sources, or weight ownership.

3. 03

   Contract constraints: The original agreement explicitly limited Microsoft from training large-scale models independently.

4. 04

   Data flywheel leakage: Enterprise fine-tuning data on OpenAI APIs may, under some terms, feed competitor improvements.

5. 05

   Iteration lag: Anthropic is on Opus 4.8 and OpenAI on GPT-5.6 while Microsoft waits on third-party releases.

The turning point came in late 2025. Both parties renegotiated. The new agreement removed model-size restrictions and explicitly allowed Microsoft to pursue superintelligence on its own. Microsoft AI chief Mustafa Suleyman put it this way:

Build 2026 was Microsoft's first public showcase of that in-house brain — and a declaration that the OpenAI-independent path has just started.

The other five MAI models: image, voice, transcribe, and code

MAI-Image-2.5 / Flash — text-to-image and image-to-image

Microsoft's first image model supporting both text-to-image and image-to-image. It ranks #2 on Arena.ai's image editing leaderboard and #3 on text-to-image. Core capabilities include Text-to-Image, Image-to-Image style transfer, and Control with Preservation (keeps semantic structure during edits). Integrated into PowerPoint and OneDrive, and listed in the Azure Foundry Model Catalog.

MAI-Transcribe-1.5 — speech-to-text

Supports 43 languages with auto-detection. FLEURS average WER 4.9% (among the lowest in the industry), Artificial Analysis WER 2.4% (3rd overall), and processing speed of 276x realtime (one hour of audio transcribed in seconds). Latency improved 5.7x vs version 1.4. The Contextual Biasing feature boosts domain-term accuracy. Pricing: $0.36 / audio hour. Beats Scribe V2, Whisper-large-V3, GPT-4o-Transcribe, and Gemini 3.1 Flash on benchmarks. Typical use cases: Teams meeting notes, customer-service transcription, Copilot voice input.

MAI-Voice-2 — multilingual TTS

• Zero-shot voice cloning: Clone a speaker from seconds of reference audio
• Emotional style control: Adjustable tone, pace, and emotional color
• Language coverage: 15+ newly added languages
• Output: MP3 at 24 kHz; pricing $22 / 1M characters
• Flash variant: Ultra-low-latency version for real-time voice agents, coming soon

Integrated into Azure Foundry, VS Code, Dynamics 365, and Microsoft Copilot.

MAI-Code-1-Flash — coding assistant (generally available now)

An inference-efficient coding model optimized for GitHub Copilot and VS Code — running in your IDE today. 256K context window, built into GitHub Copilot (including CLI), VS Code, and GitHub Actions. Pricing: $0.75 / 1M input tokens, $4.5 / 1M output tokens. SWE-Bench score 51%, beating Claude Haiku 4.5 with clear speed and cost advantages. FrontierNews.ai called it the MAI model with the most direct daily impact on developers.

All 7 models: availability overview

Developer access guide: six-step runbook and citable hard data

Six-step access runbook

1. 01

   Confirm available models: MAI-Code-1-Flash, Image-2.5, Transcribe-1.5, and Voice-2 are generally available; Thinking-1 requires a private preview application.

2. 02

   Enable Azure Foundry: Visit ai.azure.com and search the Model Catalog for MAI models.

3. 03

   Apply for MAI-Thinking-1 preview: Search "MAI-Thinking-1" in the Model Catalog and click Apply, or visit microsoft.ai/models/mai-thinking-1.

4. 04

   GitHub Copilot users: MAI-Code-1-Flash is already a Copilot backend model (CLI and VS Code inline suggestions) — no configuration change needed.

5. 05

   API calls: Use the Azure OpenAI-compatible interface with api_version 2026-05-01 and model set to mai-code-1-flash.

6. 06

   Multi-model coexistence: A single Foundry workspace can call both MAI models and GPT-5.6, routing by task complexity.

import openai

client = openai.AzureOpenAI(
    azure_endpoint="https://<your-resource>.openai.azure.com/",
    api_key="<your-api-key>",
    api_version="2026-05-01"
)

response = client.chat.completions.create(
    model="mai-code-1-flash",
    messages=[
        {"role": "system", "content": "You are an expert software engineer."},
        {"role": "user", "content": "Refactor this Python function to use async/await: ..."}
    ],
    max_tokens=2048
)
print(response.choices[0].message.content)

Citable hard data

• MAI-Thinking-1 active parameters: 35B (sparse MoE, ~1T total) — inference cost far below dense flagships
• MAI-Transcribe-1.5 speed: 276x realtime, $0.36/audio hour, FLEURS 43-language WER 4.9%
• Surface Dev Box compute: 1 PFLOPS, 128GB unified memory, 100W TDP, runs 120B+ models locally
• MAI-Code-1-Flash pricing: $0.75/M input + $4.5/M output, SWE-Bench 51%
• SWE-Bench Pro gap: MAI-Thinking-1 52.8% vs Opus 4.8 69.2% (~16 points)

Cloud-only MAI APIs can cut inference costs, but iOS signing chains, Xcode local builds, Metal inference, and 24/7 CI runs still need physical macOS nodes. The Surface Dev Box launches US-only this fall with no price yet, and VMs carry performance overhead plus EULA risk. For a more stable production environment suited to iOS CI/CD and AI agent automation, VpsMesh Mac Mini cloud rental is usually the better fit: physical Apple Silicon, root access, and predictable monthly pricing — complementing MAI-Code-1-Flash on the inference layer. Copilot handles codegen; a cloud Mac runs builds and signing.

Sources: Microsoft AI: MAI-Thinking-1 · Azure AI Foundry Blog · Surface RTX Spark Dev Box · The Verge · VentureBeat