CoDA-Bench: Code and Data-Intensive AI Agent Benchmark

CoDA-Bench is a benchmark created by RUC-DataLab to evaluate AI agents on code and data-intensive tasks in realistic environments. Unlike benchmarks providing oracle data directly, it requires agents to discover relevant data among hundreds of semantically similar files. The dataset was last updated on June 16, 2026.

Use Cases

• Benchmarking AI agent performance on data discovery tasks based on the requirement to find relevant files among hundreds of semantically similar ones.
• Evaluating code intelligence in realistic data-intensive environments based on the benchmark's joint focus on code and data intelligence.
• Training or fine-tuning code-capable AI models on tasks that integrate data retrieval and code generation.

Strengths

• Designed as the first benchmark to jointly evaluate code intelligence and data intelligence of AI agents.
• Simulates realistic data-intensive environments requiring agents to discover data rather than using oracle data.

Limitations

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

Provenance

Source

RUC-DataLab

Freshness

Last updated 2026-06-16 03:08:58