EC2-Bench: Systematic Analysis of Edge-Cloud Collaboration for Long-Horizon LLM Agents
NortheasternUniversity1923 · hf · 2026-10-20
Researchers from Northeastern University released EC2-Bench, a benchmark systematically evaluating edge-cloud collaboration in long-horizon LLM agent execution.
- Tasks require multi-step execution in local workspaces (email threads, calendars), e.g. scheduling a meeting: read emails, compare availability, resolve conflicts, send invites.
- Four collaboration strategies are evaluated: query routing (all-edge vs all-cloud), per-step routing, edge-controlled delegation, and edge-controlled advisory.
- The benchmark probes three questions: performance-cost-privacy trade-offs, the impact of edge model scale (2B/9B/27B/Plus), and how collaboration should shift as edge capability grows.
- Key findings: collaboration involves explicit trade-offs among performance, resource use, and privacy; stronger edge models favor more edge-controlled collaboration.
More from Infra
- 47 US AI data center projects canceled as CoreWeave's debt costs hit 15% — SumitGup · 2026-10-20
- Open-source Berth auto-swaps AI engines on a single 24GB GPU with lease-based VRAM handoff — Distinct-Pie2389 · 2026-10-20
- Self-hosted 14B hits 88% on contract review vs Claude's 90%, at 1/27 the cost — Broad_Breakfast_1172 · 2026-10-20
- Everyone's fighting over models, but energy owns the AI stack — r0ck3t23 · 2026-10-20
- Renting GPUs Is No Longer Special: Neoclouds Buy Software to Build Moats — FinanceYF5 · 2026-10-20
- Neoclouds' New Playbook: GPU Rental Is Commoditized, So They're Buying Software Moats — FinanceYF5 · 2026-10-20