Reasoning Model Scaling

Why it matters

Reasoning scaling reshapes training. Understanding shapes future.

The architecture

Pretrain gives base.

RL (verifier-based) extends reasoning.

Inference-time compute expands further.

How it works end to end

DeepSeek R1 + Kimi K1.5 + OpenAI o1 papers.

Verifier-based rewards.

Multi-axis scaling.