---
title: "Microsoft Introduces Seven AI Models: From Coding to Vulnerability Scanning at a Fraction of GPT-6 Cost"
description: "Microsoft has unveiled seven AI models: MAI Code 2, MAI Cyber 2 Flash, MAI Image 2.7, MAI Realtime 1, MAI Transcribe 2, and the Decision 1.1 router. The company claims GPT-6 and Claude-level performance at significantly lower costs."
date: 2026-11-18T01:30:08.238Z
lang: en
url: https://xab.info/en/posts/microsoft-introduces-seven-ai-models
tags: [microsoft, ai-models, machine-learning, cybersecurity, software-development]
author: "Alexander Buran"
publisher: "XAB.info"
---

# Microsoft Introduces Seven AI Models: From Coding to Vulnerability Scanning at a Fraction of GPT-6 Cost

![Cubic illuminated Microsoft logo on a dark background in modern tech design style](https://xab.info/media/2026/11/18/microsoft-mai-ai-models-lineup/microsoft-mai-ai-models-lineup-1.webp)

## Key Points

- Microsoft introduced a lineup of seven AI models for coding, cybersecurity, image generation, voice interaction, transcription, and query routing.
- MAI Code 2 delivers 98% of GPT-6.1 Sol performance at about 40% of the competitor's price, and 96% of Claude Opus 5.5 at 5% cost.
- MAI Cyber 2 Flash performs vulnerability scanning tasks at less than 2% of the cost of GPT-6 Astra; 90% of customers who started scanning continue using the tools.
- MAI Image 2.7 ranked first in the Artificial Analysis image editing quality ranking at one-fifth the cost of the competitor.
- Decision 1.1 automatically selects the optimal model for a query, operating 50 times faster and 79 times cheaper than GPT-Sol 6.1, with about 4,000 clients.

Microsoft has announced the release of a comprehensive lineup of seven proprietary AI models covering virtually all key areas of applied artificial intelligence: code generation and analysis, software vulnerability scanning, image creation and editing, voice interaction, speech recognition, and automated query routing between models. The announcement, made against the backdrop of fierce competition with OpenAI and Anthropic, highlights a strategic shift by the company: instead of chasing absolute benchmark leadership, Microsoft is focusing on the price-to-performance ratio, offering enterprise clients solutions that approach the functionality of competitor flagship models while costing a fraction of the price.

### MAI Code 2: Coding at a Substantially Lower Cost than Competitors

The centerpiece of the lineup is MAI Code 2, a model focused on writing, refining, and debugging program code. According to internal benchmarks published by Microsoft, the model delivers 98% of the performance of GPT-6.1 Sol at roughly 40% of the cost of the competing solution. Compared to Claude Opus 5.5, it achieves 96% performance while incurring only 5% of Anthropic's model cost. Testing covered typical enterprise tasks, code work, and software tools. Of particular note is a blind comparison conducted with developers: in 68% of cases, participants preferred the results of MAI Code 2 over GPT-6 Sol when creating interface elements, control panels, and visualizations. At the same time, the company is transparent about its limitations: the model is currently geared primarily toward standard programming tasks and is not optimized for fundamental scientific research or prolonged autonomous operation of AI agents. For mass repetitive operations—such as code explanation, minor edits, and bug fixes—a lightweight version, MAI Code 2 Flash, has been prepared, with performance comparable to GPT-6 Luna at 65% of the cost and Claude Haiku 5.5 at about 20%. MAI Code 2 is already available via GitHub Copilot, and the Flash version has appeared in Microsoft Foundry under limited preview.

### MAI Cyber 2 Flash: Vulnerability Scanning at Less Than 2% of Competitor Cost

A separate focus in the lineup is MAI Cyber 2 Flash, designed for software security and automated vulnerability detection. Microsoft claims that on the CyberGymE2E test, the model's quality is comparable to GPT-6 Astra, while task execution costs less than 2% of the competing solution. Compared to the previous MAI Cyber 1.1 Flash, the cost of processing output tokens has decreased by 86%. The company also reports that previous models in the MAI Cyber family already power nearly 40% of scans in the MDASH system, and 90% of customers who began using these tools for scanning have continued working with them. These figures indicate that the automated security audit segment is becoming one of the most commercially mature areas in Microsoft's portfolio.

### Images and Voice: MAI Image 2.7 and MAI Realtime 1

In the field of image processing, Microsoft introduced MAI Image 2.7 and its lightweight Flash version. MAI Image 2.7 secured first place in the Artificial Analysis ranking for image editing quality, outperforming GPT Image 2.6 Sunburst in its maximum configuration at roughly one-fifth of the cost. In text-to-image generation, the model ranked fourth overall, which Microsoft estimates corresponds to second place among AI developers. The Flash version is designed for high-volume generation, allowing the creation of about 1,000 images for roughly $10. For voice interaction, the company developed MAI Realtime 1. According to internal and external tests, including evaluations using Dynamics 365, the model outperformed GPT-Realtime 2.1. Processing costs start at $0.01 per minute of audio, making the solution accessible for a wide range of enterprise scenarios—from call centers to voice assistants.

### MAI Transcribe 2: Medical Transcription Handling 10 Million Requests Per Month

MAI Transcribe 2 is designed for speech recognition and transcription. It powers Dragon Copilot Transcribe, which is used in healthcare to document doctor-patient conversations. According to Microsoft, the system processes about 10 million such requests monthly, and the use of the model has halved the error rate in multilingual transcription. Today, approximately one-third of American doctors use the technology. This metric demonstrates that AI transcription has ceased to be an experimental feature and has become an infrastructure element of clinical practice.

### Decision 1.1 and MAI Licensed Weights: Routing and Business Adaptation

In addition to generative models, Microsoft introduced Decision 1.1, a system for automatically selecting the appropriate AI model for a specific query. Routing allows simple tasks to be directed to faster and cheaper models, leaving expensive systems for complex queries. According to the company, Decision 1.1 operates 50 times faster and costs 79 times less than GPT-Sol 6.1. It is currently used by about 4,000 clients, and in the Artificial Analysis Decision Index, Microsoft's development outperforms models from Jev, Google, and OpenAI. The company paid special attention to adapting AI for business needs: the new MAI Licensed Weights program allows organizations to obtain Microsoft base models, fine-tune them with their own data and requirements, while maintaining control over the infrastructure. Thus, Microsoft is building not just a model catalog, but an ecosystem where every task—from writing a line of code to security auditing—is handled by a specialized, economically optimized tool.

## Fact-Check & Sources

- [Microsoft представила семь ИИ-моделей: одни пишут и исправляют код, другие ищут уязвимости, редактируют изображения и общаются голосом](https://3dnews.ru/1151438) - Primary source, the text fully corresponds to the basis of the article. All figures and model names are taken from this source.
- [Microsoft представила семь ИИ-моделей: одни пишут и исправляют код, другие ищут уязвимости ...](https://3dnews.ru/1151438/microsoft-predstavila-sem-iimodeley-odni-pishut-i-ispravlyayut-kod-drugie-ishchut-uyazvimosti-redaktiruyut-izobrageniya-i-obshchayutsya-golosom) - Duplicates source 1, contains no additional facts.

## FAQ

### What is MAI Code 2 and how does it differ from GPT-6.1 Sol?

MAI Code 2 is Microsoft's model for writing and refining program code. According to the company, it delivers 98% of GPT-6.1 Sol performance at about 40% of the competitor's price. The model is available via GitHub Copilot.

### How much cheaper is MAI Cyber 2 Flash compared to competitors in vulnerability scanning?

Microsoft claims that on the CyberGymE2E test, the quality of MAI Cyber 2 Flash is comparable to GPT-6 Astra, while the task execution cost is less than 2% of the competing model.

### Where is MAI Transcribe 2 already used?

MAI Transcribe 2 powers Dragon Copilot Transcribe, used in healthcare to document doctor-patient conversations. According to Microsoft, the system processes about 10 million requests monthly, and roughly one-third of American doctors use the technology.

### What does Decision 1.1 do?

Decision 1.1 automatically selects the optimal AI model for a specific query, routing simple tasks to cheaper models and complex ones to expensive ones. Microsoft claims it operates 50 times faster and 79 times cheaper than GPT-Sol 6.1.