LLaMA 3

Meta's LLaMA 3, trained on 15 trillion tokens in 8B and 70B sizes, is now the open-weight model caught in a three-front war. Graph data shows it competes directly with GPT-4o, Gemma 4, and Claude Code—each a distinct threat. GPT-4o dominates API revenue; Gemma 4 targets mobile inference; Claude Code attacks developer tooling. LLaMA 3's dependency on Community Notes for alignment and fine-tuning for specialization reveals a reliance on external ecosystems, not proprietary data. Recent Meta moves—a $27B Louisiana data center and an Instagram shopping agent—signal infrastructure and application bets, but LLaMA 3 itself saw zero mentions in the last 7 days. Meanwhile, NVIDIA's Nemotron Ultra (550B open-weight) enters the ring. LLaMA 3's momentum is real but stalling; its moat is openness, but competitors are closing fast.

• ·LLaMA 3 competes with GPT-4o, Gemma 4, and Claude Code across API, mobile, and developer segments.
• ·Zero mentions in last 7 days despite 17 total—velocity is dropping.
• ·Relies on Community Notes (alignment) and fine-tuning (specialization), not proprietary data.
• ·Meta's $27B data center and Instagram shopping agent are infrastructure/application plays, not model updates.
• ·Nemotron Ultra's 550B open-weight entry threatens LLaMA 3's position in the open-weight space.