# 控制大语言模型中的推理努力程度

- 原标题：Controlling Reasoning Effort in LLMs
- 类型：博客
- 来源：Sebastian Raschka
- 发布时间：2026-07-18
- 首次精选：2026-10-28 16:25
- 分级：值得看
- 学习层级：进阶
- 主题：[reasoning-continual-learning](<https://oplearnx.com/md/topics/reasoning-continual-learning.md>)（第 4 位）
- 原文链接：[https://magazine.sebastianraschka.com/p/controlling-reasoning-effort-in-llms](<https://magazine.sebastianraschka.com/p/controlling-reasoning-effort-in-llms>)
- 本站页面：[https://oplearnx.com/resources/pk\_2be8751c7abf](<https://oplearnx.com/resources/pk_2be8751c7abf>)

## 摘要

Sebastian Raschka 的文章，从 OpenAI o1、DeepSeek-R1 及可验证奖励强化学习（RLVR）讲起，并以新近发布的 GPT-5.6 模型家族（三个尺寸、每个约五到六档推理努力设置）为例。文章聚焦如何开发具备多种努力模式的推理模型，区别于其此前侧重把常规 LLM 转为推理模型的资料。

## 原文涉及

- [Tulu 3: Pushing Frontiers in Open Language Model Post-Training](<https://arxiv.org/abs/2411.15124>)
- [Kimi k1.5: Scaling Reinforcement Learning with LLMs](<https://arxiv.org/abs/2501.12599>)
- [DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning](<https://arxiv.org/abs/2501.12948>)
- [Qwen3 Technical Report](<https://arxiv.org/abs/2505.09388>)
- [DeepSeekMath-V2: Towards Self-Verifiable Mathematical Reasoning](<https://arxiv.org/abs/2511.22570>)

---
来源：LearnMap · 学径（https://oplearnx.com）