LongWriter-Zero: Mastering Ultra-Long Text Generation via Reinforcement Learning

TL;DR

LongWriter-Zero introduces a reinforcement learning approach to enable ultra-long, high-quality text generation in large language models without relying on synthetic data, outperforming traditional supervised fine-tuning methods.

Contribution

This work presents a novel RL-based training method from scratch for ultra-long text generation, eliminating the need for synthetic data and achieving state-of-the-art results.

Findings

Outperforms supervised fine-tuning on long-form writing tasks.

Achieves state-of-the-art metrics on WritingBench and Arena-Write.

Surpasses larger models like DeepSeek R1 and Qwen3-235B.

Abstract

Ultra-long generation by large language models (LLMs) is a widely demanded scenario, yet it remains a significant challenge due to their maximum generation length limit and overall quality degradation as sequence length increases. Previous approaches, exemplified by LongWriter, typically rely on ''teaching'', which involves supervised fine-tuning (SFT) on synthetic long-form outputs. However, this strategy heavily depends on synthetic SFT data, which is difficult and costly to construct, often lacks coherence and consistency, and tends to be overly artificial and structurally monotonous. In this work, we propose an incentivization-based approach that, starting entirely from scratch and without relying on any annotated or synthetic data, leverages reinforcement learning (RL) to foster the emergence of ultra-long, high-quality text generation capabilities in LLMs. We perform RL training starting from a base model, similar to R1-Zero, guiding it to engage in reasoning that facilitates planning and refinement during the writing process. To support this, we employ specialized reward models that steer the LLM towards improved length control, writing quality, and structural formatting. Experimental evaluations show that our LongWriter-Zero model, trained from Qwen2.5-32B, consistently outperforms traditional SFT methods on long-form writing tasks, achieving state-of-the-art results across all metrics on WritingBench and Arena-Write, and even surpassing 100B+ models such as DeepSeek R1 and Qwen3-235B. We open-source our data and model checkpoints under https://huggingface.co/THU-KEG/LongWriter-Zero-32B