Microsoft unveiled a new family of AI models specialized in coding, cybersecurity and image generation, seeking to compete with Anthropic and OpenAI at lower costs. According to the company, MAI Code 2 delivers 96% of Claude Opus 5.5’s performance on enterprise coding tasks for 5% of its price, while MAI Cyber 2 Flash completed 92.3% of vulnerability-discovery tasks against 91.5% for GPT-6 Astra, at less than 2% of the cost per task. MAI Image 2.7 led an independent image-editing ranking with 1,215 points, compared with 1,211 for GPT Image 2.6 Sunburst, and costs about $35 per 1,000 images. These results have limitations, however: the coding comparisons include a promotional discount, the image ranking’s uncertainty ranges overlap, and Code 2 is not yet optimized for fundamental scientific research or long-running autonomous assignments.
Source: Read the original article


