LLM Performance on Synthetic Household Data: Distributional Fit and Generation Time

Michael Jones authored a benchmark comparing 9 large language models on an 800-household test case. The dataset, last updated on June 2, 2026, records distributional fit, structural feasibility, and generation time metrics. It is a small, 5.5 KB Excel file shared under a CC-BY-4.0 license.

Use Cases

• Benchmarking LLM generation speed based on reported generation time metrics.
• Comparing the structural feasibility of synthetic outputs across different LLM architectures.
• Evaluating the distributional fidelity of LLM-generated synthetic household records.
• Selecting an LLM for synthetic data generation tasks based on a multi-criteria performance score.

Strengths

• Directly compares 9 distinct LLMs on a common task.
• Includes three specific performance metrics: distributional fit, structural feasibility, and generation time.
• Uses a defined test case of 800 households for consistent evaluation.
• Shared under a permissive CC-BY-4.0 license.

Limitations

• The dataset is very small at 5.5 KB, indicating limited scope.
• Row count and column-level documentation are unknown, requiring manual inspection after download.
• The description metadata is limited; actual data quality and methodology details require verification.

Provenance

Source

Michael Jones via figshare.

Collection Method

Likely generated by running 9 LLMs on a synthetic household data generation task.

Freshness

Last updated 2026-06-02 17:34:01; freshness should be verified.