The AI Model Accessibility Checker (aimac.ai)

The AI Model Accessibility Checker (AIMAC) has been launched to evaluate how well various AI models generate accessible web pages, addressing a critical issue as 95.9% of the top million websites fail basic accessibility standards. By prompting leading AI models across 28 categories to create web pages, AIMAC allows developers to see side-by-side comparisons of accessibility performance, highlighting the necessity for AI in coding to focus on inclusivity rather than replicating human coding mistakes. OpenAI's models dominated the leaderboard, with GPT 5.4 Mini and GPT 5.3 Codex leading with zero accessibility violations, while significant improvements were noted in Google's Gemini 3.1 Pro Preview, which rose from last to eighth place. This benchmark is particularly significant as it emphasizes the growing role of AI in shaping the web's accessibility landscape. The analysis indicates that while some models, like those from OpenAI and Alibaba's Qwen, manage to produce more accessible code efficiently, others like Anthropic's Claude models lag behind, raising concerns about their broader claims of intelligence and responsibility. The results suggest that AI developers must prioritize accessibility from the outset to harness AI's potential in creating a more inclusive web, aligning with the AI/ML community's goals of advancing technology for all users.