Microsoft Unveils New AI Models at Ignite Conference
Microsoft introduced a new lineup of specialized AI models at its Ignite conference, highlighting cost-efficiency and performance benchmarks against competitors.
Microsoft unveiled a new lineup of specialized AI models at its Ignite conference on Tuesday. The announcement includes MAI Code 2, which delivers 96% of Claude Opus 5.5's coding performance at 5% of its cost. It also achieves 98% of GPT-6.1 Sol's performance at 40% of the price. MAI Code 2 Flash targets routine fixes and quick edits, offering performance comparable to Claude Haiku 5.5 at 20% of the price. Additionally, MAI Code 2 Flash provides comparable performance to GPT-6 Luna at 65% of the cost.
In the cybersecurity sector, MAI Cyber 2 Flash succeeded on 92.3% of CyberGym-E2E's evaluated vulnerability-discovery tasks. This model costs less than 2% of the cost per task compared to GPT-6 Astra. For visual tasks, MAI Image 2.7 topped an independent image-editing leaderboard with 1,215 points, surpassing GPT Image 2.6 Sunburst's 1,211 points. Microsoft quotes about $35 per 1,000 images for this model.
MAI Code 2 and Code 2 Flash are available in GitHub Copilot. Code 2 Flash is in limited-access preview in Microsoft Foundry. Microsoft also optimized a single model, MAI Work Flash, to serve multiple products across M365 and Fabric IQ. The MAI model for Excel is on par for accuracy and token efficiency with GPT 6 Luna. The MAI model for PowerPoint performs on par with GPT 6 Luna in both quality and token efficiency. The MAI model for Word uses approximately 42% fewer input tokens per call than GPT6.1 Sol. The MAI model for Teams beats GPT 6 Luna at tracking decisions, filling knowledge gaps, and adapting custom instructions. The MAI model for FabricIQ outperforms GPT-5.6-sol, GPT-6-astra, and Opus-5.5 at DAX query generation.
Sources
- Hacker News: front pageMAI Code 2 and Code 2 Flash ↗
- DecryptWhy Microsoft Thinks Smaller, Cheaper AI Can Beat Claude and GPT ↗
Every sentence of this story was checked against facts quoted from these articles. How we make the news
Spotted an error? Tell us
