Tier 3 uses on-demand Firecracker microVMs with under 2 second P99 latency and VM-level isolation for ML training, GPU workloads, and system dependencies so that the heaviest workloads run in the strongest isolation tier.

| Tier 3 | On-demand Firecracker microVM | N/A | <2s | VM-level | ML training, GPU workloads, system dependencies |

— PYRANA_PLATFORM_ARCHITECTURE_REFERENCE.md

### Three-Tier Execution Model

— PYRANA_PLATFORM_ARCHITECTURE_REFERENCE.md

Tier 3

— PYRANA_PLATFORM_ARCHITECTURE_REFERENCE.md