LetTokenBurnEdge–Cloud Collaborative Agent Benchmark
A benchmark evaluating edge-cloud collaboration in long-horizon LLM agent execution across task performance, resource usage, and privacy exposure using digital-workspace tasks.
$ npm i -g lettokenburn-cli$ ltb bench info ec2-bench
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| GLM-5.3 | 82.7 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| DeepSeek V4 Pro | 72.9 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| Qwen3.7 Plus | 71.3 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| qwen3.5-27B | 61.8 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| qwen3.5-9B | 39.1 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| Qwen3.5 2B | 12.8 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
EC2-Bench provides executable digital-workspace agent tasks within a unified framework to assess edge-cloud collaboration strategies. It evaluates systems based on task success, computational resource consumption, and the extent of sensitive data exposure to the cloud.
The benchmark facilitates systematic analysis of how edge capability impacts collaboration efficiency. Key findings indicate that collaboration balances performance and privacy better than edge-only or cloud-only approaches, with privacy exposure remaining a key differentiator as edge models improve.