Microsoft introduced new AI models designed for programming, cybersecurity, and image generation during the Ignite 2026 conference.
On November 17, Microsoft launched a series of AI models designed for programming, vulnerability detection, and image generation during its Ignite 2026 conference. The company claims these new models deliver performance comparable to Claude and GPT at significantly reduced costs.
The standout offerings include MAI Code 2 and its lighter variant, Code 2 Flash. MAI Code 2 reportedly achieves 96% of Claude Opus 5.5's performance for only 5% of its cost. Meanwhile, it reaches 98% of GPT-6.1 Sol's results at just 40% of the expense, according to Microsoft’s benchmarks across four metrics.
These benchmarks, which assess practical programming tasks such as debugging and repository management, were conducted on Microsoft's MSBench platform, with one using data from an independent service, Artificial Analysis.
MAI Code 2 is tailored for everyday corporate tasks such as debugging and application preparation, while Code 2 Flash is optimized for brief operations and frequent queries, offering quality that rivals Claude Haiku 5.5 at 20% of the cost and GPT-6 Luna at 65%.
While the pricing for both models was calculated with a promotional discount of 50%, Microsoft acknowledged that the models have limitations, particularly in handling complex scientific research and long-term autonomous tasks requiring multi-step planning. Both models are now available via GitHub Copilot, with Code 2 Flash undergoing limited pilot testing at Microsoft Foundry.
New Cybersecurity Model Outperforms Competitors
A key addition to Microsoft’s AI lineup is MAI Cyber 2 Flash, aimed at identifying vulnerabilities and analyzing threats. In the CyberGym-E2E benchmark, it successfully completed 92.3% of tasks compared to 91.5% for GPT-6 Astra, with task costs under 2% of its competitor’s expenses.
MAI Cyber 2 Flash employs a contextual window of 1 million tokens, quadrupling the capacity of its predecessor, and will gradually replace earlier models in the MDASH platform starting in mid-December.
Advancements in Image Editing and Voice Interaction
Microsoft also announced the release of MAI Image 2.7 and its lighter version, Image 2.7 Flash, which excelled in independent rankings for image editing. MAI Image 2.7 scored 1215 Elo points, surpassing its main competitor, GPT Image 2.6 Sunburst, which scored 1211.
The cost of MAI Image 2.7 is approximately $35 for 1,000 images, while the Flash version is around $10. The company claims these prices are significantly cheaper than those of its competitors at maximum quality settings.
Additionally, Microsoft introduced new solutions for real-time voice interaction, transcription, and automatic AI model selection, contributing to its expanding AI ecosystem.