# Should I use Llama 4 Scout API or self-hosted with Zilliz Cloud?

Source: https://zilliz.com/ai-faq/should-i-use-llama-4-scout-api-or-selfhosted-with-zilliz-cloud
Author: zilliz

API simplicity vs. self-hosted cost control—API is easier for startups; self-hosted wins on ROI after ~10M tokens/month at enterprise scale.

API access (via Meta or providers) is simple: send query, pay per token, no infrastructure. Self-hosting requires deployment overhead but zero per-token costs. With Zilliz Cloud managing retrieval (fully managed), adding self-hosted Scout is operationally reasonable: Zilliz scales automatically, you only manage Scout. For enterprises running millions of tokens/month through RAG, self-hosting usually pays for itself. For smaller teams, API access is preferred until scale justifies infrastructure.

Hybrid approach: use Zilliz Cloud (no-ops managed retrieval) + Scout API (pay-as-you-go generation) to minimize upfront infrastructure. Then migrate to self-hosted Scout once volume justifies it. Both approaches work equally with Zilliz Cloud; the difference is operational burden vs. cost optimization.

---
**Related Resources**
- [Zilliz Cloud — Managed Vector Database](https://zilliz.com/what-is-zilliz-cloud) — managed retrieval layer
- [Retrieval-Augmented Generation (RAG)](https://zilliz.com/learn/Retrieval-Augmented-Generation) — hybrid architectures
- [Zilliz Cloud Pricing](https://zilliz.com/pricing) — vector storage cost calculator