Benchmark RadarAI BENCHMARK PROFILE
ROUTEBENCH
ROUTEBENCH evaluates whether transformers can learn latent algorithm routing across four solver families (ridge, lasso, Huber, kNN) in a controlled diagnostic setting.
- Released
- 2026-07-27
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
Provides controlled evidence on internal routing in transformers, but lacks a standalone public comparison path and is primarily a research probe.
Motivation
A central question in the in-context learning literature is whether transformers can organize episode-level adaptation around different inductive-bias families.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.