Meta releases Llama-3x: multimodal instruction model optimized for UX research

AI · 6 min read

Llama-3x is Meta's latest multimodal instruction model, designed specifically for product and UX teams. The model can simulate realistic user personas, generate moderated interview transcripts, and produce synthesis artifacts like journey maps with explicit reasoning traces.

A key addition is a privacy filter layer that redacts personally identifiable information and annotates sensitive content, addressing common concerns about running research transcripts through large models. The model also supports fine-grained role prompts so researchers can request outputs from a target persona's perspective.

Meta is releasing Llama-3x under a tiered license with free access for academic and non-commercial research. Enterprise tooling includes audit logs and retrainable privacy filters to meet corporate compliance needs.