Benchmark RadarAI BENCHMARK PROFILE
SkillReason-Bench
SkillReason-Bench is a retrieval benchmark with 3,729 queries and 61,228 skills across nine domains, but the paper focuses on proposing a retrieval method (SkillReason) rather than the benchmark itself.
- Released
- 2026-08-09
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
The benchmark serves as a testbed for the proposed method and is compared to existing benchmarks, but the primary contribution is the method, not a standalone reusable benchmark.
Motivation
Large language model agents increasingly rely on reusable skills to extend their capabilities beyond parametric knowl- edge.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.