Хочете будувати кар’єру в Data Science та AI? Vodafone Big Data Lab відкриває новий набір —>
Nataliia KhandusenkoAI Eng
18 November 2026, 08:20
2026-11-18
Microsoft introduces AI models for programming and cybersecurity
Microsoft Corp. announced a new line of artificial intelligence models with enhanced programming and cybersecurity capabilities at its Microsoft Ignite conference in San Francisco on Tuesday, Nov. 17. The company is stepping up competition with rivals such as Anthropic and OpenAI.
Microsoft Corp. announced a new line of artificial intelligence models with enhanced programming and cybersecurity capabilities at its Microsoft Ignite conference in San Francisco on Tuesday, Nov. 17. The company is stepping up competition with rivals such as Anthropic and OpenAI.
The seven-model Microsoft (MAI) line includes the 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.
Cybersecurity
The MAI Cyber 2 Flash model is designed for continuous cybersecurity monitoring and protection, in particular, for detecting and remediating vulnerabilities in real-world systems, writes Interesting Engineering. Microsoft claims that in the CyberGymE2E test it provides quality at the level of GPT-6 Astra, but costs less than 2% of its price for the task. The new model is based on previous MAI Cyber systems, which account for almost 40% of checks on the MDASH security assessment platform (and 90% of checks started with MAI were continued with it).
Programming
For software developers, the company has released MAI Code 2, aimed at everyday enterprise coding tasks. According to Microsoft, the model delivers 98% of the performance of GPT-6.1 Sol at about 40% of the cost, compared to Anthropic’s Claude Opus 5.5, which delivers 96% of the performance at just 5% of the price. A lighter version of MAI Code 2 Flash is designed for routine tasks like code explanation, point-and-click edits, and bug fixes; it’s already available on GitHub Copilot and in limited preview via Microsoft Foundry.
Image, speech and routing
MAI Image 2.7: An image processing model that ranks first in image editing and fourth in image generation from text in the Artificial Analysis rankings. Microsoft claims it outperforms OpenAI GPT Image 2.6 Sunburst while costing five times less.
MAI Realtime 1: The company's first low-latency speech-to-speech model, starting at $0.01 per minute of audio.
Dragon Copilot Transcribe: A medical transcription tool based on MAI Transcribe 2, serving 10 million patient appointments each month and halving errors in multilingual transcription.
Decision 1.1: An application request routing system that routes tasks to large models only when needed. The model is 50x faster and 79x cheaper than GPT-Sol 6.1, and is already in use by 4,000 customers.
Customization and custom chips
The company also introduced MAI Licensed Weights, which allows businesses to tune models on their own sensitive data and run them in their own Azure environments. Microsoft also highlighted its second-generation AI accelerator, the Maia 200, which is now in production, powers the company’s largest models and delivers up to 1.7x more tokens per watt than leading graphics processing units (GPUs).
Mustafa Suleiman, head of the AI division at Microsoft, said in an interview with Reuters that the strategy of creating highly specialized technologies for individual industries allows us to keep costs low while keeping up with competitors, and in some tests, even surpassing them.
"This is not just one model or a short-term flash. This is a system—a continuous improvement mechanism that we have built to simultaneously optimize performance and price," Suleiman emphasized.
He said recent cybersecurity incidents — including the hacking of the Hugging Face platform by OpenAI agents — have forced him to focus not only on the power of systems but also on critical safeguards. Suleiman named a top priority for himself a research area known as alignment — preventing malicious actions by artificial intelligence and ensuring that systems behave exactly as intended.
Як нейромережі бачать вільну та незалежну Україну? Тест dev.ua
Нейронні мережі для генерації зображень бачать світ по-своєму, їхню логіку зрозуміти часом зовсім неможливо. Але таки хочеться. На честь Дня Незалежності України редакція dev.ua вирішила провести невеликий експеримент.
Ми задали чотирьом різним нейронним мережам п’ять однакових запитів: «прапор України», «День Незалежності України», «український Крим», «перемога України» та «українці». Отриманими результатами ми ділимося з вами нижче.
У TikTok тепер можна генерувати фон за допомогою нейромережі. Ми протестували її та ділимося результатами
У TikTok з’явилася нова функція «Розумний фон». З її допомогою як фон для тіктоків можна підставляти згенеровані нейромережею зображення. Редакція dev.ua протестувала цю технологію і ділиться своїми враженнями.