Meta unveils Llama-3 Code, a 1.2T parameter model tuned for software and UI generation

AI · 6 min read

Llama-3 Code brings scale and targeted fine-tuning to code-heavy and UI-focused prompts, reducing hallucinations in component APIs and layout constraints.

Meta included specialized training on component libraries, CSS frameworks, and cross-platform UI patterns so that designers and engineers can generate scaffolded interfaces consistent with chosen frameworks. The model supports constrained generation modes that prioritize existing component inventories.

Meta also rolled out a developer console for inspecting token-level attributions and provenance so teams can audit why a particular implementation was suggested and how to correct it.

Availability starts with limited enterprise access and a permissive research license for universities. Meta states the release aims to accelerate design-to-development handoffs while preserving reproducibility and compliance.