From shared model access to dedicated GPU compute. Every plan includes all 136 models with automatic routing and failover.