parcs-benchmark/parcs-agent-benchmark

---
license: mit
task_categories:
- other
tags:
- parcs
- benchmark
- parallel-computing
- ai-agents
pretty_name: "PARCS-Agent Benchmark — 15 Parallel Computing Tasks"
---

# PARCS-Agent Benchmark

15 tasks for evaluating PARCS-enabled (parallel) vs. sequential AI agents.

## Main split — `benchmark`

| Column | Type | Description |
|--------|------|-------------|
| `task_id` | int | Task number 1–15 |
| `question` | string | Self-contained prompt given to the agent |
| `answer` | string | JSON with reference / ground-truth values |

## Supplementary splits

Seed-based tasks include pre-generated input data:
`task_03_cities`, `task_04_sir_reference`, `task_05_data`, `task_05_configs`,
`task_06_patients`, `task_07_sequences`, `task_08_train`, `task_08_test`,
`task_10_portfolio`, `task_10_scenarios`, `task_11_library`,
`task_13_nodes`, `task_13_edges`, `task_13_sources`, `task_15_sequences`

## Citation

PARCS-Agent benchmark (Bohusevych, 2026). Forthcoming.