Learning path · Models & Tokens · 12
Open vs Closed Models
Closed models are proprietary APIs; open-weight models can be self-hosted, fine-tuned, and inspected—each with different ops and compliance tradeoffs.
Why it matters
- Data residency and air-gapped deployment often require open weights.
- Closed models may lead on frontier reasoning with less MLOps burden.
- Hybrid stacks route sensitive workloads locally and frontier tasks to APIs.
Key ideas
- Self-hosting
- API convenience
- License constraints
Closed models offer fast iteration, strong defaults, and provider-managed safety—but you accept vendor terms, network egress, and opaque version changes. Open models give control over weights, fine-tuning, and deployment geography at the cost of GPU ops, quantization tuning, and safety tooling you must assemble. Many enterprises use both: open models inside the VPC for PII-heavy extraction, closed models for complex reasoning with redacted inputs. Negotiate contracts covering training opt-out, retention windows, and incident notification before routing regulated payloads to any closed API.
Updated 2026-08-09 · Full learning path