Benchmark RadarAI BENCHMARK PROFILE
IsoSci
IsoSci evaluates reasoning vs. knowledge retrieval in LLMs using isomorphic cross-domain problem pairs, enabling controlled attribution of reasoning gains.
- Released
- 2026-07-01
- Readiness
- Inspectable
- Primary field
- General AI
Why it matters
Provides a method to separate reasoning ability from knowledge, challenging assumptions about chain-of-thought and guiding model evaluation.
Motivation
We introduce ISOSCI, a benchmark of isomorphic cross-domain science problem pairs that separates reasoning ability from domain knowledge retrieval in LLM evaluation.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.