Microsoft has introduced seven new MAI models designed to deliver ultimate performance at significantly lower prices. These new AI models will provide budget-friendly tools across cybersecurity, coding, image generation, audio transcription, and industry-level decision-making. Some applications require AI agents to run 24/7, which can increase operational costs and directly affect the budget. Providing high-performance MAI models at the lowest cost can help organizations run AI models 24/7 without worrying much about skyrocketing costs. This article highlights key features of the newly launched world-class MAI models.

Microsoft introduces 7 new MAI models at scalable prices
AI models process prompts or instructions by breaking them into small fragments, called tokens. AI agents use these tokens to read commands or instructions, generate responses, and perform tasks. Automated tasks require AI agents to run 24/7, which significantly increases token consumption and costs. Microsoft’s new MAI models can help address this challenge. Microsoft claims that these MAI models will deliver exceptional results at the lowest costs.
Key highlights of the MAI models
MAI Cyber 2 Flash is designed to deliver frontier intelligence in the field of cybersecurity. It is built on earlier MAI Cyber models used by most cybersecurity experts. Microsoft claims that MAI Cyber 2 Flash beats GPT -6 Astra in terms of quality on the CyberGym-E2E benchmark test at less than 2% of the cost per task. This latest model is priced 86% lower on output tokens than MAI Cyber 1.1 Flash.
MAI Code 2, an AI agent built for everyday enterprise coding tasks, delivers 98% of the coding quality of GPT-6.1 Sol at about 40% of the price. Comparing it against Opus 5.5, it delivers 96% performance at just 5% of price. MAI Code 2 Flash also shows outstanding performance against GPT-6 Luna at 65% of the price, and Haiku 5.5 at 20% of the price.
Microsoft’s MAI Image 2.7 has become the best image-editing AI model in the world, securing the first position in artificial analysis. It outperformed GPT Image 2.6 Sunburst (max) at just 1/5th of the price. In text-to-image generation, MAI Image 2.7 ranked number 4, making Microsoft the second-best lab.
Additionally, MAI Image 2.7 Flash offers world-class image generation at approximately $10 per thousand images.
MAI Realtime 1 is the first live speech-to-speech model, designed for natural and low-latency conversations. Microsoft claims that it delivers better performance than GPT Realtime 2.1 and starts at just $0.01 per audio minute.
Dragon Copilot Transcribe, built on MAI Transcribe 2, is the world’s fastest and most accurate multilingual transcription model. It is currently used by 1/3 of US physicians and handles about 10 million sessions monthly.
Decision 1.1 is the leading decision model in the market. It provides model routing, so that you can use pricier models only when you really need them. It is currently used by 4000 customers and is 50 times faster and 79 times cheaper than GPT-6.1 Sol.
Maia 200, Microsoft’s second-generation AI accelerator, is now in production running our largest models, achieving 1.7x more tokens per watt
These new MAI models will provide developers and organizations with high-performing AI tools at a lower cost. You can read the detailed information in Microsoft’s official blog post.