EC2-Bench: edge-cloud collaboration hits 73% of cloud-only agent performance at 15% of the cost
zibuyu9 · x · 2026-10-25
An OpenBMB-led study on arXiv systematically analyzes edge-cloud collaboration in long-horizon LLM agent execution, introducing EC2-Bench: 128 executable tasks evaluated jointly on task completion, resource cost and privacy exposure across four collaboration strategies.
Key findings:
- Small cloud budget, big gains: with Qwen3.5-9B on-device and GLM-5.3 in the cloud, edge-only scores 39.12; the best collaboration reaches 60.46 — about 73% of cloud-only (82.70) — at $8.73, roughly 15% of cloud-only's $56.42
- Performance converges, privacy doesn't: on Qwen3.7-Plus, strategy gaps shrink to 3.67 completion points, yet privacy exposure spans 66.23 percentage points — and the best-performing strategy leaks the most
- As edge models strengthen, collaboration approaches cloud-only performance, but deciding when the cloud should execute vs. just advise remains a critical design choice
More from coding & agent
- After a week under active cyberattack, an engineer explains why software factories are now essential — danielrock · 2026-10-25
- Bots talking to bots: one agent auto-contacts Guo Yu, whose bot answers on his behalf — huangyun_122 · 2026-10-25
- Free open-source AI engineering course builds every algorithm from raw math before frameworks — ghumare64 · 2026-10-25
- "No human will ever write code better than LLMs," declares engineer Arpit Bhayani — arpit_bhayani · 2026-10-25
- Dev launches staygreen.io, a lifetime-free uptime monitor, after balking at $330/mo white-label pricing — PawelHuryn · 2026-10-25
- ComfyUI plugin turns Load Image into a mini Photoshop with editable OpenPose — Occsan · 2026-10-25