Back

Microsoft unveils cheaper in-house AI models for coding, security, images, speech and routing
SiTech AI Team3 min read

Microsoft unveils cheaper in-house AI models for coding, security, images, speech and routing

Microsoft AI unveiled seven specialized models at Ignite for coding, cybersecurity, image editing, speech, routing and retrieval, positioning them as cheaper alternatives to OpenAI and Anthropic systems for always-on agents.

Microsoft AI unveiled seven specialized models at Ignite, covering coding, cybersecurity, image editing, speech, routing and retrieval. The company positions them as cheaper alternatives to OpenAI and Anthropic systems for always-on AI agents. Several are already available through GitHub Copilot, Foundry and other platforms, while others remain in preview or exclusive programs.

Coding models

MAI Code 2 averages 98 percent of GPT-6.1 Sol's performance across four benchmarks at 40 percent of the cost, and 96 percent of Claude Opus 5.5's performance at 5 percent of the cost, Microsoft says. MAI Code 2 Flash targets code explanations, targeted edits and routine fixes, matching GPT-6 Luna and Claude Haiku 5.5 at 65 and 20 percent of their respective costs. Microsoft derived MAI Work Flash from the Flash model; it runs in Microsoft 365 and Fabric IQ, matches GPT-6 Luna on internal Excel and PowerPoint benchmarks, performs similarly to GPT-6.1 Sol in Word while using about 42 percent fewer input tokens, and outperforms GPT-5.6 Sol, GPT-6 Astra and Opus 5.5 on DAX queries in Power BI. In GitHub Copilot, at the promotional price that applies through May 17, 2027, MAI Code 2 costs $0.50 per million input tokens and $2.50 per million output tokens, then $1 and $5; MAI Code 2 Flash costs $0.05 and $0.25 until then, rising to $0.10 and $0.50.

Security and images

MAI Cyber 2 Flash solves 92.3 percent of tasks in CyberGym-E2E, compared with 91.5 percent for GPT-6 Astra, which needs twice the reasoning budget, Microsoft says. Per task, the MAI model costs less than 2 percent of GPT-6 Astra, with per-token costs at one hundredth; on WebExploitBench it scores 40.4 percent versus 38.4 percent for GPT-5.6 Cyber. MAI Image 2.7 leads the Artificial Analysis image-editing leaderboard with 1,215 Elo points from 5,478 votes, four points ahead of GPT Image 2.6 Sunburst (max) at one-fifth the cost. It ranks fourth for text-to-image generation and costs about $35 per 1,000 images, while Flash costs about $10. The model can change color, material or position of individual objects without altering the rest of the image and generates transparent backgrounds.

Speech, routing and retrieval

MAI Realtime 1 is Microsoft's first speech-to-speech model, designed for interruptions, waiting until speakers finish and recognition with background noise. It outperforms GPT Realtime 2.1 on Big Bench Audio, τ-Voice and Dynamics 365 tests, Microsoft says, and is compatible with the OpenAI Realtime API. Microsoft charges one US cent per audio minute, with reasoning extra at $1 per million input tokens and $5 per million output tokens. Microsoft-Decision-1.1 handles routing, classification, prioritization and workflow control; built on Google's Gemma 4 26B A4B, it scores 68 points on Decision Index v2.1, six more than its predecessor, and performs as accurately as a large language model but 2.2 times faster and 4.4 times cheaper. Microsoft-Retriever-1 gathers information from enterprise documents for agents in Foundry IQ and launches as a preview. Through MAI Licensed Weights, companies can fine-tune MAI models with their own data and run them in their own Azure environment; one example is AT&T, where a customized model for 5G network operations analysis reportedly outperformed GPT-6.1 Sol on standard tasks at roughly one-fiftieth of the cost.

Sources: The Decoder

SiTech

SiTech — AI-powered web development

We build fast, modern websites and bring AI into real business workflows. Have a project or a question? We'd love to help.