2,500 Concurrent Sandboxes Per Node, Cross-Scaffold Checkpoint Merging Yield Free Gains

nrehiew_ · x · 2026-09-11

nrehiew explains the sandbox infra: a custom orchestrator places sandboxes across compute nodes with local bottleneck checks, each VM in a NUMA domain supports 2,500 concurrent sandboxes with priority scheduling, and hacking triggers a "repercussion" signal. Agent trajectories are for the first time explicitly inspired by real-world partner usage patterns, and merging checkpoints across different scaffolds and configs delivered free performance gains.

Related event: DeepSeek V4.1 Tech Report Deep Dive: RL Infrastructure, Sandbox Design and Inference Stack(8 posts)→

Original post →

More from Infra

Infra channel →