---
title: "Microsoft Unveils Seven New MAI AI Models: From Cybersecurity to Voice, at 50x Lower Cost Than GPT"
description: "At the Ignite conference, Microsoft introduced seven specialized MAI AI models—from cybersecurity to voice. The company states they are 5–79 times cheaper than OpenAI and Anthropic equivalents at comparable quality."
date: 2026-11-18T08:24:29.928Z
lang: en
url: https://xab.info/en/posts/microsoft-launches-seven-specialized-mai-models
tags: [microsoft, ai, mai-models, ignite-2026, openai, anthropic, machine-learning, cybersecurity, azure]
author: "Alexander Buran"
publisher: "XAB.info"
---

# Microsoft Unveils Seven New MAI AI Models: From Cybersecurity to Voice, at 50x Lower Cost Than GPT

![Display showing seven new Microsoft MAI AI models at a digital conference booth](https://xab.info/media/2026/11/18/microsoft-mai-ai-modeli-desh-le-gpt-claude/microsoft-mai-ai-modeli-desh-le-gpt-claude-1.webp)

## Key Points

- Microsoft unveiled seven specialized MAI AI models at Ignite instead of one universal system
- MAI Cyber 2 Flash rivals GPT-6 Astra on CyberGymE2E but costs less than 2% of its price per task
- MAI Code 2 offers 98% of GPT-6.1 Sol performance at ~40% cost, and 96% of Claude Opus 5.5 efficiency at 5% price
- MAI Realtime 1 is the company's first speech-to-speech model, starting at $0.01 per minute
- Decision 1.1 operates 50 times faster and 79 times cheaper than GPT-Sol 6.1
- Production is powered by the Maia 200 AI accelerator, generating 1.7 times more tokens per watt

At the annual Ignite conference, Microsoft announced the launch of seven new specialized neural networks under the MAI line. The tech giant's strategy differs fundamentally from its competitors: instead of building a single universal model for all tasks, the company has opted for narrowly focused solutions designed for specific workflows. As reported by RBC-Ukraine with reference to Interesting Engineering, the key competitive advantage of these new models is pricing — in all stated scenarios, they significantly outperform counterparts from OpenAI and Anthropic, which should enable companies to scale autonomous AI agents without a sharp increase in token costs.

### Cybersecurity and Development: Flagship MAI Models

The first in the line is MAI Cyber 2 Flash, a model designed for continuous cybersecurity monitoring, vulnerability detection, and remediation in enterprise systems. Microsoft claims that in CyberGymE2E tests, it rivals the quality of GPT-6 Astra, yet the cost per task is less than 2 percent of the competitor's price. For daily coding tasks, MAI Code 2 is available: according to the company, the model provides 98 percent of GPT-6.1 Sol performance at about 40 percent of its cost. Compared to Anthropic’s Claude Opus 5.5, the model delivers 96 percent efficiency at 5 percent of the price. A lightweight version, MAI Code 2 Flash, is focused on routine operations—code explanation, quick targeted fixes—and is already available in GitHub Copilot.

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

Updates also include multimodal and voice solutions. MAI Image 2.7 creates and edits images; according to Microsoft, it topped the Artificial Analysis photo editing tools ranking, surpassing OpenAI's GPT Image 2.6 Sunburst, while being five times cheaper to use. MAI Realtime 1 is the company's first model for direct speech-to-speech interaction with minimal latency. Usage costs start at $0.01 per minute of audio, a figure that makes voice AI assistants economically viable for mass corporate adoption.

### Medicine and Optimization: Dragon Copilot Transcribe and Decision 1.1

For the medical industry, Microsoft introduced Dragon Copilot Transcribe, a speech transcription tool based on MAI Transcribe 2. According to the company, the system already handles 10 million patient encounters monthly and has halved the error rate in multilingual transcription. Another distinct product is Decision 1.1, an intelligent router that redirects complex tasks to more powerful models only when strictly necessary. Microsoft claims the solution is 50 times faster and 79 times cheaper than GPT-Sol 6.1, making it an efficient 'dispatcher' for multi-level AI pipelines.

### Infrastructure: MAI Licensed Weights and the Maia 200 Chip

Alongside software products, the corporation unveiled MAI Licensed Weights, which allow businesses to fine-tune neural networks on their own confidential data and deploy them in a private Azure cloud environment. To power the largest models in production, Microsoft uses its second-generation AI accelerator, the Maia 200. The company estimates the chip generates 1.7 times more tokens per watt of energy consumed compared to leading GPUs on the market, reducing operational infrastructure costs and the carbon footprint of data centers.

The MAI lineup presented at Ignite reflects a shift in the industry from the race for absolute quality in universal models to the economics of specialized solutions. If Microsoft's claimed performance and cost metrics are confirmed by independent benchmarks, this could significantly alter the corporate AI market landscape and accelerate the adoption of autonomous agents in sectors where token costs were previously a major barrier.

## Fact-Check & Sources

- [Microsoft бросает вызов GPT и Claude: за ИИ теперь можно платить в 50 раз меньше](https://www.rbc.ua/ukr/news/microsoft-kidae-viklik-gpt-i-claude-shi-teper-1794989377.html) - The announcement at Ignite is confirmed. All quantitative figures (98%, 96%, 50x, 79x, 1.7x) are statements by Microsoft and were not verified by independent tests at the time of publication.

## FAQ

### How many MAI models did Microsoft introduce at Ignite?

Seven: MAI Cyber 2 Flash, MAI Code 2, MAI Code 2 Flash, MAI Image 2.7, MAI Realtime 1, Dragon Copilot Transcribe, and Decision 1.1. Additionally, MAI Licensed Weights and the Maia 200 chip were presented.

### How does MAI differ from the OpenAI and Anthropic approach?

Microsoft has focused on narrowly specialized models for specific tasks rather than one universal system. This reduces token costs for every specific use case.

### How much cheaper are MAI models compared to competitors?

According to Microsoft: MAI Cyber 2 Flash is less than 2% of GPT-6 Astra's price; MAI Code 2 is 40% of GPT-6.1 Sol cost and 5% of Claude Opus 5.5 price; MAI Image 2.7 is 5 times cheaper than GPT Image 2.6 Sunburst; Decision 1.1 is 79 times cheaper than GPT-Sol 6.1.

### What is MAI Licensed Weights?

These are licensed model weights that allow companies to fine-tune neural networks on their own confidential data and deploy them in a private Azure cloud environment.

### Which chip does Microsoft use to run MAI models?

The company's own second-generation AI accelerator, the Maia 200. It is estimated to generate 1.7 times more tokens per watt compared to leading GPUs on the market.