NLCO: Benchmark for LLMs on Natural-Language Combinatorial Optimization

NLCO is a benchmark dataset containing 6,450 total samples across 43 tasks for evaluating large language models on natural-language combinatorial optimization problems. The dataset, created by summer142857jiang and last updated in April 2026, is organized into 129 CSV files with 50 samples per file and three difficulty tiers: Set-S, Set-M, and Set-L.

Use Cases

• Benchmarking LLM performance on combinatorial optimization tasks based on the 43 described tasks.
• Evaluating LLM reasoning across difficulty tiers based on the described Set-S, Set-M, and Set-L buckets.
• Analyzing model performance on constrained problem-solving based on the dataset's focus on a 'constrained world'.
• Comparing LLM capabilities on structured reasoning tasks based on the benchmark's evaluation design.

Strengths

• Contains 6,450 total samples for evaluation.
• Provides 43 distinct reasoning tasks.
• Organizes problems into three defined difficulty tiers (Set-S, Set-M, Set-L).

Limitations

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

Provenance

Source

huggingface user summer142857jiang

Collection Method

Likely created as a normalized release for benchmarking, as described.

Time Range

null

Freshness

Last updated 2026-04 06:58:26; freshness should be verified.

Geography

null