Inkling-Small Model Update: Continuous RL Significantly Boosts Coding Capabilities

LiTianleli · x · 2026-07-31

Inkling-Small began with on-policy distillation from the main Inkling model, achieving initial results comparable to the original. The team then ran two weeks of continuous Reinforcement Learning (RL) specifically focused on coding and agentic capabilities.

The author notes that this continuous training strategy produced an even stronger final model, exclaiming 'RL goes brrrrr!'

Original post →

More from Models

Models channel →