JIUTIAN-TReB: A Multi-Dimensional Table Reasoning Benchmark with 7,790 Test Cases

JT-LM created the TReB dataset to evaluate large language models on table reasoning, comprehension, and processing. It contains 7,790 high-quality test cases spanning a spectrum from fundamental language understanding to advanced data analysis. The dataset was last updated on June 17, 2026.

Use Cases

• Benchmarking table reasoning performance based on the described 6 core skills and 26 subtasks.
• Evaluating model comprehension of hierarchical data structures based on the dataset's multi-dimensional and hierarchical design.
• Testing language models on fundamental language understanding tasks related to tables.
• Assessing advanced data analysis capabilities using the high-quality test cases.

Strengths

• Contains 7,790 high-quality test cases.
• Covers a complete capability spectrum with 6 core skills and 26 subtasks.
• Designed as a multi-dimensional and hierarchical evaluation benchmark.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

JT-LM via Hugging Face.

Collection Method

Likely constructed for research and evaluation purposes, as described in the associated paper.

Freshness

Last updated 2026-06-17 07:38:47; freshness should be verified.