BixBench: A Revised Benchmark for Question Answering

BixBench is a dataset for evaluating question-answering systems, created by the author 'futurehouse'. The dataset was substantially revised in September 2025 to address insufficient detail in original questions. It was last updated on October 1, 2025.

Use Cases

• Benchmarking question-answering models based on the revised open-answer questions.
• Evaluating model performance on questions with varying levels of detail.
• Training models to handle questions that have been reviewed for answerability.

Strengths

• Dataset underwent a substantial revision in September 2025 to improve question answerability.
• Last update was recent, on 2025-10-01 02:07:44.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The full description is hosted externally, requiring a click-through for complete details.

Provenance

Source

futurehouse on Hugging Face

Freshness

Last updated 2025-10-01 02:07:44.