Anthropic unveils Claude-XM: multimodal model focused on design reasoning

AI · 4 min read

Claude-XM is Anthropic’s latest multimodal offering, trained to perform critique, generate heuristic-based suggestions, and output structured JSON for design-tool integration. The company claims it can analyze screenshots and return prioritized UX issues, suggesting fixes with estimated implementation effort.

A unique selling point is Claude-XM’s “design rationale” output: for each suggestion it provides a short human-readable justification and cites best-practice heuristics or WCAG criteria where applicable. This aims to make automated audits more actionable for designers and engineers.

Anthropic is releasing SDKs that map Claude-XM’s structured outputs to popular design platforms and issue trackers. Early partners include accessibility firms and large design ops teams who plan to integrate the model into sprint workflows for continuous UX validation.

Privacy and evaluation were highlighted: Anthropic said Claude-XM underwent iterations with subject-matter experts and external audits to reduce hallucination in critique mode. The model is available via API and will roll out to enterprise customers with customizable safety and verbosity controls.