Friday, November 20, 2026
AI desk
/
/
Microsoft Launches Seven MAI Models for Coding, Images and Voice

Microsoft Launches Seven MAI Models for Coding, Images and Voice

Microsoft launched seven MAI models for coding, cybersecurity, images, voice, transcription and decision-making on November 17, 2026.
Last updated
November 20, 2026
7 min read
Fact-checked

Photo: TechJournal

Share

Quick Answer

Microsoft has launched seven MAI models for coding, cybersecurity, image generation, real-time voice, transcription, and decision-making. The November 17, 2026, release puts several models in GitHub Copilot, Microsoft Foundry, or MAI Labs. Microsoft reports performance gains, but its comparisons use selected benchmarks and can include promotional discounts. Developers should verify availability, access limits, and pricing before choosing a model.

Key Takeaways

  • Microsoft announced 7 MAI models on November 17, 2026.
  • MAI Code 2 is available in GitHub Copilot for coding work.
  • MAI Cyber 2 Flash is initially limited to a select customer group.
  • MAI Realtime 1 starts at $0.01 per audio minute through MAI Labs.
  • Microsoft’s benchmark and price comparisons need independent validation.

What did Microsoft launch with its new MAI models?

Microsoft launched a family of 7 MAI models covering cybersecurity, coding, image generation, real-time voice, transcription, and decision-making. The company announced the models during Ignite on November 17, 2026, as part of a broader effort to place Microsoft-developed AI systems across its developer tools and business products. Microsoft’s Ignite announcement describes the release as a set of specialized models rather than one general-purpose assistant.

The Microsoft MAI models matter because Microsoft is assigning different systems to different tasks, including code generation, security analysis, image editing, voice conversations, and routing decisions. A specialized approach can make deployment more efficient when a smaller or faster model handles a narrow workload. The limitation is that Microsoft has not presented the family as a single replacement for every existing AI model or service.

Microsoft’s launch also arrives as developers compare several proprietary model ecosystems for coding and workplace automation. Readers evaluating Microsoft’s approach can also consider how new AI models for ChatGPT work and coding are being positioned for similar enterprise tasks, although the products, access methods, and model claims differ.

Which Microsoft MAI models are available now?

Microsoft MAI model availability depends on the individual model and the service where Microsoft has placed it. MAI Code 2 is available in GitHub Copilot, while MAI Code 2 Flash is available in GitHub Copilot and in limited-access preview through Microsoft Foundry. MAI Image 2.7 is available in Microsoft Foundry, and MAI Realtime 1 is available to try through MAI Labs.

MAI Cyber 2 Flash has a more restricted rollout. The cybersecurity model is initially available to a limited group of customers, according to a Reuters report on the launch. That access limit matters because a public announcement does not mean every Copilot, Foundry, or Azure customer can deploy each model immediately.

Microsoft-Decision-1.1 is already integrated into Copilot routing and Cowork’s Auto mode, according to Microsoft. The practical result is that some users may interact with the decision model indirectly, through Microsoft’s product routing, rather than selecting it manually in a model picker.

ModelPrimary purposeAvailability stated by Microsoft
MAI Code 2CodingGitHub Copilot
MAI Code 2 FlashHigh-volume coding tasksGitHub Copilot and limited Microsoft Foundry preview
MAI Cyber 2 FlashCybersecurity workInitially limited customer access
MAI Image 2.7Image generation and editingMicrosoft Foundry
MAI Realtime 1Speech-to-speech interactionsMAI Labs
Microsoft-Decision-1.1Model routing and decisionsCopilot routing and Cowork Auto mode

How does MAI Code 2 compare for coding work?

MAI Code 2 is Microsoft’s coding model for GitHub Copilot, and Microsoft says it reaches 98% of GPT-6.1 Sol’s coding quality at about 40% of the price across 4 coding benchmarks. Microsoft’s comparison is self-reported, so the figures describe the company’s own evaluation rather than an independent market-wide test. The claim is most relevant to teams that measure coding tools by both output quality and cost.

MAI Code 2 Flash is designed for high-volume work, including code explanations, targeted edits, and routine fixes. Microsoft says the faster model is available in GitHub Copilot and as a limited-access Microsoft Foundry preview. Developers handling repetitive maintenance tasks may value lower latency and throughput, while complex architectural changes still require code review, testing, and human accountability.

AI coding tools can accelerate routine work, but generated code remains a security and reliability risk when it is accepted without review. Teams should use tests, dependency checks, access controls, and established secure-development practices. Organizations concerned about AI-generated attack planning should note that government testing of major AI systems has found that models can produce viable cyberattack plans under some conditions.

What does MAI Cyber 2 Flash change for security teams?

MAI Cyber 2 Flash is Microsoft’s new cybersecurity model, and Microsoft says output tokens cost 86% less than MAI Cyber 1.1 Flash. Microsoft also says MAI Cyber 2 Flash beats GPT-6 Astra on CyberGym-E2E at less than 2% of the cost per task. Both comparisons are self-reported vendor claims, so security teams should treat them as early performance and pricing signals rather than final procurement evidence.

The cost claim matters because cybersecurity workflows can involve large volumes of alerts, logs, code, and incident documentation. Lower output-token costs can make an AI assistant more practical for repeated analysis, but cheaper output does not prove that a system is safe to use with sensitive enterprise data. Security teams still need to confirm data handling terms, retention controls, identity permissions, and audit logging before submitting incident materials.

MAI Cyber 2 Flash also has limited initial access, which means the model is not yet a standard option for every organization. The most sensible approach is to test a restricted deployment with nonproduction data and clear human review rules. Security teams should stop and consult their Microsoft account representative or internal security leadership before connecting an AI model to production systems, privileged repositories, or live incident-response workflows.

How do MAI Image 2.7 and MAI Realtime 1 work?

MAI Image 2.7 is Microsoft’s image-generation and image-editing model in Microsoft Foundry. Microsoft says MAI Image 2.7 leads Artificial Analysis’s image-editing ranking and costs one-fifth as much as GPT Image 2.6 Sunburst (max). The ranking and cost comparison are Microsoft’s claims, and users should validate output quality, edit consistency, and applicable pricing for their own workloads.

MAI Image 2.7 may be useful for product imagery, marketing drafts, and editing workflows where teams need generated visuals alongside existing assets. Image models still require careful review because a convincing output can contain inaccurate details, altered labels, or unwanted visual artifacts. Teams producing product shots should verify glass, reflections, and labels, which remain important concerns in AI image editing workflows.

MAI Realtime 1 is Microsoft’s speech-to-speech model, and Microsoft says access starts at $0.01 per audio minute through MAI Labs. Real-time voice systems can support natural spoken interactions without requiring a separate text-only exchange. The practical limitation is that voice data can be sensitive, so organizations should confirm who can access recordings, transcripts, and any derived data before using the model for customer or employee conversations.

What does Microsoft-Decision-1.1 do in Copilot?

Microsoft-Decision-1.1 is a decision-making model integrated into Copilot routing and Cowork’s Auto mode. Microsoft says the model is used by 4,000 customers and is 50 times faster and 79 times cheaper than GPT-6.1 Sol. Those speed and cost comparisons are self-reported, but they show Microsoft’s focus on using a lower-cost model to select or route work inside larger AI systems.

Model routing matters because an AI product does not always send every request to the same underlying system. A routing model can determine which model handles a request based on the task, which may affect response speed, capability, and cost. Users may not see that decision process directly, so administrators should review available product documentation and enterprise controls before treating Auto mode as predictable for regulated or high-risk work.

Microsoft-Decision-1.1 is not a substitute for organizational judgment. A fast routing decision can reduce infrastructure costs, but it cannot determine whether an AI response is legally accurate, secure, or appropriate for a customer-facing workflow. Organizations should keep human approval requirements for consequential decisions involving employment, finance, health, legal matters, or system access.

How reliable are Microsoft’s benchmark and price claims?

Microsoft’s performance and cost figures are useful launch information, but they are not independent proof that each MAI model is the best option for every customer. Microsoft compares MAI Code 2 with GPT-6.1 Sol across 4 coding benchmarks, while MAI Cyber 2 Flash is compared with GPT-6 Astra on CyberGym-E2E. Benchmark selection, prompts, scoring rules, model versions, and workloads can materially affect the result.

Microsoft’s published comparisons also rely on selected benchmarks, and some stated price advantages include promotional discounts, Axios reported in its launch analysis. That qualification matters because promotional pricing and benchmark conditions may not match a customer’s long-term production costs or internal tasks.

The practical response is to run a controlled evaluation before moving a workload. Organizations should compare output quality, latency, failure rates, operational costs, data controls, and human-review requirements using representative tasks. A model that performs well on a benchmark can still be a poor match for an organization’s codebase, security policy, or customer data restrictions.

What should developers do before using Microsoft MAI models?

Developers should begin with a narrow, reversible pilot before moving Microsoft MAI models into a production workflow. Start with non-sensitive sample tasks, define success criteria, and compare results against an existing process. This approach helps teams separate a promising launch claim from measurable improvements in their own environment.

  1. Identify one task, such as code explanation, image editing, or voice interaction, with a clear quality standard.
  2. Confirm the model’s current access path through GitHub Copilot, Microsoft Foundry, MAI Labs, or the relevant Microsoft product.
  3. Review pricing, promotional terms, usage limits, and data-handling controls before sending organizational data.
  4. Test outputs against representative examples and record errors, review time, and operating costs.
  5. Require human approval before deploying generated code, security recommendations, or customer-facing content.

Microsoft MAI models can be useful where their specialized design matches the task, but availability and commercial terms vary by model. Teams should stop before connecting a model to production data or privileged systems if they cannot verify access controls, retention practices, and escalation procedures. A vendor representative, security team, or legal and compliance lead should review high-risk deployments before launch.

FAQ

What are Microsoft MAI models?

Microsoft MAI models are a family of 7 AI models for coding, cybersecurity, image generation, real-time voice, transcription, and decision-making. Microsoft announced the family on November 17, 2026, during Ignite.

Is MAI Code 2 available in GitHub Copilot?

MAI Code 2 is available in GitHub Copilot. Microsoft says MAI Code 2 reaches 98% of GPT-6.1 Sol’s coding quality at about 40% of the price across 4 coding benchmarks, a self-reported comparison.

Who can use MAI Cyber 2 Flash?

MAI Cyber 2 Flash is initially available to a limited customer group. Security teams should verify eligibility and data controls before testing the model with organizational information.

How much does MAI Realtime 1 cost?

MAI Realtime 1 starts at $0.01 per audio minute through MAI Labs. Organizations should confirm current pricing and applicable usage terms before deploying voice features at scale.

Are Microsoft’s MAI benchmark claims independently verified?

Microsoft’s MAI benchmark claims are not independent verification because Microsoft published the comparisons. Selected benchmarks and promotional discounts can affect reported performance and pricing advantages.

Share this guide
Facebook
X
LinkedIn
Written by
James Chen is a technology journalist covering artificial intelligence, software tools, and the future of work. He has been testing and reviewing AI products since 2023 and has hands-on experience with every major AI platform. His work focuses on helping everyday users get more done with AI — without the hype.

In this article

The AI Brief

Guides like this, every Friday.

One email. No hype cycle.

Essential guides