Thai Legal Benchmark For Retrieval-Augmented Generation

NitiBench provides test data for evaluating large language model frameworks like RAG and LCLM on Thai legal text. The benchmark, created by VISAI-AI and last updated in November 2025, consists of two distinct datasets: NitiBench-CCL and NitiBench-Tax.

Use Cases

• Benchmarking RAG system accuracy on Thai legal queries using the separated reasoning and final answer fields.
• Evaluating LCLM performance on the NitiBench-Tax dataset for tax-related legal question answering.
• Fine-tuning language models for Thai legal text generation using the preprocessed question-answer pairs.
• Testing model reasoning capabilities on the NitiBench-CCL dataset derived from the WangchanX-Legal-ThaiCCL-RAG source.

Strengths

• Dataset includes two specialized legal sub-domains: CCL and Tax.
• Preprocessing step separates reasoning process from final answer for clearer evaluation.
• Last updated in November 2025, indicating recent maintenance.

Limitations

• Unknown total row count and dataset size limits reproducibility assessment.
• Specific column names and data structure are not provided in the input.
• Geographic scope is limited to Thai legal text, reducing generalizability.

Provenance

Source

Derived from the WangchanX-Legal-ThaiCCL-RAG Dataset.

Collection Method

Includes an additional preprocessing step to separate reasoning from final answer.

Time Range

null

Freshness

Last updated 2025-11 26.

Geography

Thailand