TY  - RPRT
TI  - Do Instruction-Tuned Models Always Perform Better Than Base Models? Evidence from Math and Domain-Shifted Benchmarks
AU  - Prateek Munjal
AU  - Clement Christophe
AU  - Ronnie Rajan
AU  - Praveenkumar Kanithi
PY  - 2026
UR  - https://arxiv.org/abs/2601.13244
ID  - 2601.13244
ER  -