SGI-Bench: Scientist-Aligned LLM Benchmark Across 10 Disciplines

A benchmark for evaluating Scientific General Intelligence across the full inquiry cycle, spanning 10 disciplines and more than 1,000 expert-curated samples inspired by Science's 125 Big Questions. The dataset, SGI-Reasoning-Lite, was created by InternScience and last updated on 2026-06-02. It employs an agentic evaluation framework for probing LLMs.

Use Cases

• Benchmarking LLM performance on scientific reasoning tasks based on the described Deliberation, Conception, Action, and Perception inquiry cycle
• Evaluating agentic AI systems using the described scientist-aligned, multi-disciplinary framework
• Training or fine-tuning models for scientific question-answering based on the 1,000+ expert-curated samples
• Analyzing model capabilities across 10 different scientific disciplines as mentioned in the description

Strengths

• More than 1,000 expert-curated samples provide a substantial evaluation corpus
• Covers 10 distinct scientific disciplines for broad assessment
• Structured around Science's 125 Big Questions, providing a high-level conceptual foundation

Limitations

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

Provenance

Source

InternScience via Hugging Face

Collection Method

Expert-curated samples inspired by Science's 125 Big Questions

Freshness

Last updated 2026-06-02 12:05:27