Guozhen AIGlobal AI field notes and model intelligence

Realtime AI News

Microsoft ships seven in-house MAI models at Ignite to replace outside models with cheaper alternatives

Microsoft unveiled seven in-house MAI models at Ignite spanning code, security, images, speech and routing, and is moving its own products onto them at a fraction of outside-model prices. Microsoft says MAI Code 2 reaches 98 percent of GPT-6.1 Sol's performance at 40 percent of the cost, framing the launch around token economics rather than benchmark leadership.

Published
微软Ignite一次发七款MAI自研模型,要用更便宜的内部替代品换掉外部模型
Image source: microsoft.com

Microsoft's AI division used its Ignite conference to unveil seven in-house models covering code, security, images, speech and routing, shipping them in a single batch as the company moves its own products away from outside models and toward cheaper internal alternatives.

The pitch is price-to-performance rather than raw frontier capability, and Microsoft is measuring itself against its own partner's lineup. The company says MAI Code 2 averages 98 percent of GPT-6.1 Sol's performance across four benchmarks at 40 percent of the cost, and reaches 96 percent of Claude Opus 5.5 at 5 percent of the cost. Those tests are DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas and FrontierCode v1.1; three ran on Microsoft's internal MSBench platform through the GitHub Copilot CLI without sub-agents.

Pricing follows the same logic. In GitHub Copilot, MAI Code 2 costs $0.50 per million input tokens and $2.50 per million output tokens at a promotional rate through May 17, 2027, then $1 and $5. MAI Code 2 Flash runs $0.05 and $0.25, rising to $0.10 and $0.50.

The security model makes the cost argument sharpest. Microsoft says MAI Cyber 2 Flash solves 92.3 percent of tasks on the CyberGym-E2E vulnerability benchmark, against 91.5 percent for GPT-6 Astra, which needs twice the reasoning budget. Per task the MAI model costs less than 2 percent of GPT-6 Astra, at $0.45 versus $34.38 on CyberGym-E2E, with per-token prices at one hundredth. Its context window spans one million tokens, four times that of MAI Cyber 1.1 Flash. Microsoft also disclosed that about 10 percent of the roughly 900 CyberGym tasks drew on the same open-source projects and vulnerability reports as parts of its own training data, and that it removed those tasks for all models even though reference solutions never entered training.

Microsoft-Decision-1.1 handles routing and workflow control, built on Google's Gemma 4 26B A4B mixture-of-experts foundation. It scores 68 on Artificial Analysis' Decision Index v2.1, six points above its predecessor and ahead of Jev 1.14, Gemini Decision 1 and GPT-6 Luna. In Copilot it decides whether a request goes to Copilot or Cowork; Microsoft says it matches a large language model's accuracy while running 2.2 times faster and 4.4 times cheaper, at $0.042 per million input tokens with output tokens free.

The rest of the batch fills product gaps rather than chasing leaderboard peaks. MAI Image 2.7 leads the Artificial Analysis image-editing leaderboard at 1,215 Elo, four points ahead of GPT Image 2.6 Sunburst at one-fifth the cost, and can move or recolor individual objects without touching the rest of the frame. MAI Realtime 1 is Microsoft's first speech-to-speech model and is compatible with the OpenAI Realtime API, so existing voice apps can switch with relatively little rework. MAI Retriever-1 arrives as a preview in Foundry IQ.

Microsoft also pushed into licensing. MAI Licensed Weights let companies fine-tune MAI models on their own data and run them inside their own Azure environment, with data and customizations staying with the customer. As an example, Microsoft says a customized 5G network operations model at AT&T beat GPT-6.1 Sol on standard tasks at roughly one-fiftieth of the cost, though no launch date or pricing has been announced and the offering is set to arrive through the Frontier Company program.

The subtext is the April restructuring of the Microsoft-OpenAI partnership: OpenAI can now sell through any cloud, Azure remains the primary partner, and Microsoft holds a non-exclusive license to OpenAI's models and products through 2032. Every routine enterprise workload that shifts to an MAI model is leverage Microsoft no longer hands to a supplier.

The claims still come with edges. Most comparisons rest on Microsoft's own benchmarks and promotional pricing, agentic benchmark results with tools are pending, and a comparison against GPT-6 Cyber is still missing. What to watch is whether enterprises accept the 98 percent framing, and how MAI Cyber 2 Flash performs inside MDASH once it replaces the existing model for eligible customers from mid-December.

Why it matters

If enterprises buy the pitch, competition shifts from leaderboard scores to the cost of running agents all day, and Microsoft keeps more value inside its own stack instead of paying a supplier.

MicrosoftMAIGitHub Copilot
Back to realtime news

Nearby Updates

All

11/19, 03:03

OpenAI says its models delivered up to 14% gains on AMD's flagship GPU in under five months

OpenAI says its models delivered up to 14% gains in power, performance and area on AMD's flagship GPU IP in under five months, work it describes as roughly a year of chip-design progress. AMD has already folded some changes into a design sent for manufacturing, while OpenAI's new GPT-6.2 Astra led its RISC-V design evaluations.

11/19, 02:04

Google gives AI Mode in Search its own incognito-style temporary chats

Google has added temporary threads to AI Mode in Search, letting people hold conversations that are not saved to search history or used to shape future recommendations. The feature brings an incognito-style option to generative search, addressing concerns about what conversational queries leave behind.

11/19, 02:03

Google opens its Spark AI agent to free users under the name “Tasks”

Google has opened its Spark AI agent to free users, presenting the capability under the name “Tasks,” according to a report from The Tech Buzz. The shift extends agentic, multi-step assistance beyond paying tiers to a much broader base of ordinary accounts.

11/19, 02:01

Sustainalytics flags high ESG risk at Anthropic, putting AI governance in focus

A Sustainalytics assessment carried by Morningstar places Anthropic in a high ESG risk category, tying the rating to questions about AI governance. The move shows how conventional investment analysis is starting to evaluate model developers on how they govern their own technology.