Git Good Bench: 900 Samples for AI Agent Performance on Git Tasks

GitGoodBench Lite is a subset of 900 samples for evaluating AI agents in resolving Git tasks. The dataset is evenly split across Python, Java, and Kotlin programming languages and the sample types merge conflict resolution and file-commit gram. It was created by JetBrains and last updated on November 18, 2025.

Use Cases

• Benchmarking AI agent performance on Git merge conflict resolution based on the described sample type.
• Evaluating AI agent capabilities for file-commit operations based on the described sample type.
• Training or fine-tuning models for automated Git operations across Python, Java, and Kotlin codebases based on the described language split.

Strengths

• Contains 900 total samples, providing a defined scale for evaluation.
• Samples are evenly balanced with 150 per combination of the three programming languages and two task types.
• Data is sourced from 479 unique, open-source GitHub repositories, suggesting real-world provenance.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is known but the total number of features or the specific data structure is unknown.

Provenance

Source

JetBrains via Hugging Face.

Collection Method

Collected from 479 unique, open-source GitHub repositories.

Time Range

null

Freshness

Last updated 2025-11-18 09:05:25; freshness should be verified.

Geography

null