Alibaba Cloud Details AgentSandbox: 100K Sandboxes/min Created, TCO Cut Up to 70%
机器之心 · wechat · 2026-09-28
Jiqizhixin dives into Alibaba Cloud's AgentSandbox, which positions sandboxes as the execution layer of Agent Infra—serving both Agentic RL/Eval (mass parallel, resettable environments) and Agent Serving (fast startup, secure execution, task resumption).
Key specs: 100K sandbox creations per minute, million-scale concurrent sandboxes per region, cold start P99 <180ms on pre-warmed templates, hot start P99 <20ms. MicroVM isolation, per-task identity, and network policies form the security boundary. Memory-level checkpoints enable Reset/Fork/Replay, and hibernation-wakeup (P99 <600ms) releases compute during long waits. Three pricing tiers billed per second, with TCO reductions up to 70% in typical scenarios. E2B-compatible Developer API plus Kubernetes API for platform teams. MiniMax spins up 5,000 sandboxes in 10s (TCO -25%); Kimi scales to tens of thousands of sandboxes per minute.
More from coding & agent
- Google Research open-sources PaperOrchestra, a multi-agent framework for writing AI papers — google-research · 2026-11-16
- ffmpeg-skill gives coding agents local video editing via 42 FFmpeg tools and MCP — kajisho5 · 2026-11-16
- Agents Now Reply to Slack Messages on Their Own, Resolving a Prod Incident Ghost-to-Ghost — iamrobotbear · 2026-11-16
- Engineer's Interview Prep Method: Use Claude to Find Which Answers Undersell Your AI Work — brandon_galang · 2026-11-16
- DeepSeek open-sources its app frontend "chat-demo", already at 3.9k stars — NielsRogge · 2026-11-16
- lemire: AI's real impact on coding is expanding the engineer's 'cognitive radius', not speed — lemire · 2026-11-16