Microsoft Launches Seven MAI Models at Ignite, Pricing Directly Against OpenAI and Anthropic

Microsoft used its Ignite 2026 conference in San Francisco to unveil a new family of in-house AI models spanning coding, image generation, security, real-time speech, and routing — marking its most ambitious push yet to build a parallel AI supply chain alongside its partnership with OpenAI.
The company announced at least seven new models under the MAI brand, several of which went live immediately in GitHub Copilot and Azure AI Foundry. The benchmarks it cited place these models in direct competition with offerings from OpenAI and Anthropic, frequently at a fraction of the cost.
Coding Models Now Live in GitHub Copilot
MAI Code 2, Microsoft's flagship agentic coding model, is designed for complex, multi-file and repository-level engineering work. It is now available across all GitHub Copilot tiers — Pro, Pro+, Max, Business, and Enterprise — and in limited preview on Azure AI Foundry. Microsoft claims it delivers 98% of GPT-6.1 Sol's coding quality at roughly 40% of the cost, and 96% of Claude Opus 5.5's performance at 5% of the price. Those figures are averages across four benchmarks: DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas, and FrontierCode v1.1.
MAI Code 2 Flash, the lighter variant, targets high-volume, fast-turnaround tasks — repetitive fixes, quick edits, and sub-agent calls. Microsoft says it matches GPT-6 Luna at 65% of the cost. Code 2 Flash is now available to all GitHub Copilot users, including free and student accounts.
Image Model Tops Editing Benchmark at One-Fifth the Cost
MAI Image 2.7 ranked first on Artificial Analysis's AA-Image-Editing v2.1 benchmark, beating GPT Image 2.6 Sunburst at approximately one-fifth of the per-image price (around $0.035 per image). A Flash variant — MAI Image 2.7 Flash — is priced at roughly $10 per 1,000 images, making it one of the cheapest capable image generation options on the market. Both models are in public preview on Azure AI Foundry and through third-party platforms including fal, Genspark, OpenRouter, and Vercel.
Security and Real-Time Speech Round Out the Family
MAI Cyber 2 Flash is Microsoft's new security-focused model and the most capable in the MAI family by its own security benchmarks — scoring 92.3% on CyberGym-E2E versus 91.5% for GPT-6 Astra. It costs under 2% per task compared to general-purpose frontier models and carries a one-million-token context window, priced at $0.10 per million input tokens and $0.50 per million output tokens. The model is currently exclusive to Microsoft security products and replaces the existing model in its MDASH threat-analysis platform for eligible customers from mid-December 2026.
MAI Realtime 1 is Microsoft's first live speech-to-speech model, enabling sub-second, human-like voice conversations at $0.01 per audio minute. It is not yet generally available; organizations can register interest through MAI Labs. MAI Transcribe 2, which powers Dragon Copilot Transcribe, already handles 10 million clinical encounters per month.
The Bigger Picture: Microsoft AI as a Model Supplier
The wave of announcements signals that Microsoft is building out MAI as a standalone AI model business, not just an OpenAI reseller. The company introduced MAI Licensed Weights, which lets enterprise customers fine-tune a frontier MAI model on their own data and run it in their Azure environment. Its routing model, Microsoft-Decision-1.1, is now generally available at $0.042 per million input tokens and routes requests to larger models only when necessary — a cost-control mechanism with 4,000 enterprise customers already using it.
At the same time, Microsoft announced GPT 6.2 is coming to Azure AI Foundry through its ongoing OpenAI partnership, reinforcing that the two supplier tracks coexist rather than compete head-on. As reported by multiple outlets covering the Ignite keynote, Microsoft's stated goal is to give developers the full range: its own lower-cost models when price-performance is the priority, and OpenAI's latest when maximum capability matters.
Originally reported by Microsoft. Read the original article for additional details.
View original source