Verdict
Meta's 2025 Llama 4 flagship MoE (400B / 17B active). Intelligence 14 — no longer frontier, still a widely deployed open-weight production model with a Llama Community license.
“Still useful; superseded on quality by 2026 open-weight labs.”
Keep if already integrated; evaluate Kimi K3 or DeepSeek V4 for new work.
Training data policy: N/A — open weights. Self-host for full control.
DPDP Act (India): Self-host in India for full DPDP compliance.
Data residency: Self-hosted
Self-host in India