// ∞ Reasoning & Test-Time Compute
DeepSeek-R1 Reasoning RL Recipe
Large-scale outcome-reward RL alone teaches long reasoning and self-reflection.
DeepSeek-R1 推理强化学习配方 2025 Advanced core
IN PLAIN WORDS
R1-Zero applies GRPO with rule-based rewards directly to a base model, and longer thinking and self-checking emerge; R1 adds cold-start data and multi-stage training, then distills reasoning into small models.
Used in: DeepSeek-R1, its distilled models and many open replications
TIMELINE · prerequisites → this → what builds on it
01Original paper
DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
DeepSeek-AI, Daya Guo, Dejian Yang, et al. · 2025
arXiv ↗ ✓ Verified · arXiv API
02Where it sits
Learn the left column first; the right column builds on this one.
Learn first
Then you can learn
- Nothing in the atlas builds on it yet
Related:Test-Time Compute ScalingKnowledge Distillation
Left: prerequisites · right: builds on this · dashed: related. Full map →
03Learning resources
Official docs, explainers, courses and reference code. All links checked.
Reference code
Explainers
04Where the course teaches it
On learning paths:Post-training & Reasoning Models
For AI / data
- Verification: verified, checked 2026-10-21.
- JSON: /api/learn/algorithms.json (id:
deepseek-r1)