Meta AI (FAIR)

Meta’s AI strategy is unusually explicit about trade secrets versus publication: competitive inference economics paired with periodic open-weight drops that reset baseline capabilities for startups and nation-states alike. The analytical task is to separate “open” from “supportable,” and to watch safety and misuse externalities that do not show up in PR charts.

We catalog shipped Llama generations and connect them to claims about openness, safety, and the durability of Meta’s model-led moat.

Tracked claims attributed

• Despite its expensive price tag, the technology is nowhere near where it needs to be in order to be … [open]

Models shipped

• Llama 3 — *Meta’s 2024 open-weights family that re-baselined permissive licensing competition for mid-scale models.*
• Llama 4 — *Later Meta open-weights tranche extending context and multimodal baselines under the Llama brand.*