Reasoning & Test-Time Compute · 2025

Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?

Yang Yue, Gao Huang

Shows via pass@k that RLVR mainly sharpens sampling of solutions already reachable by the base model rather than expanding its reasoning boundary, reframing what current reasoning-RL actually buys.

Editorial record

Plain-language summary

Shows via pass@k that RLVR mainly sharpens sampling of solutions already reachable by the base model rather than expanding its reasoning boundary, reframing what current reasoning-RL actually buys.

Knowledge graph

Relationships

Descendants

Source record

Provenance

Record ID
P-642
Record created
2026-07-13
Last reviewed
2026-07-14
Record version
2

Citation caveat: Citation metadata is approximate and marked unverified in the source dataset.