aiqbench analysis
This paper introduces RL with Model-rewarded Thinking (RLMT), a training method that extends long chain-of-thought reasoning from rule-based domains like math to open-ended tasks such as creative writing, significantly outperforming standard RLHF pipelines. By optimizing against preference-based reward models rather than just verifiable answers, the approach demonstrates that extensive internal reasoning can be generalized for broader conversational capabilities without requiring massive pre-training datasets. This advancement rethinks post-training strategies and suggests a more efficient path to improving general-purpose chat model performance across diverse tasks.
Language Models that Think, Chat Better
arXiv:2509.20357v2 Announce Type: replace Abstract: Reinforcement learning with verifiable rewards (RLVR) trains language models to use long chain-of-thought reasoning (CoT) in domains like mathematics and code with rule-based verifiers. However, long CoT learned through RLVR does not generalize well to open-ended tasks -- such as writing essay outlines or making meal plans -- where humans reason routinely. This paper establishes the benefits of long CoT for general-purpose chat capabilities and introduces RL with Model-rewarded Thinking (RLMT)1, which pushes RLVR beyond verifiable domains. Using diverse real-world prompts, RLMT requires LMs to generate long CoT reasoning before responding, and optimizes them with online RL against a preference-based reward model used in RLHF. Across 40 training runs on Llama-3.1-8B and Qwen-2.5-7B (both base and instruct) and multiple optimization algorithms (DPO, PPO, and GRPO), RLMT consistently outperforms standard RLHF pipelines. This includes substantial gains of 3-7 points on three chat benchmarks (AlpacaEval2, WildBench, and ArenaHardV2), along with 1-3 point improvements on other tasks like creative writing and general knowledge. RLMT can also be applied directly to base models without an SFT stage, akin to DeepSeek-R1-Zero. Remarkably, with only 7K prompts, Llama-3.1-8B base trained with our RLMT recipe outperforms Llama-3.1-8B-Instruct post-trained with a complex multi-staged pipeline with 25M+ examples. We close with qualitative and quantitative analyses of how trained models plan their responses. Our results rethink the post-training pipeline and call upon future work to understand and employ thinking more broadly.
Abstract:Reinforcement learning with verifiable rewards (RLVR) trains language models to use long chain-of-thought reasoning (CoT) in domains like mathematics and code with rule-based verifiers. However, long CoT learned through RLVR does not generalize well to open-ended tasks -- such as writing essay outlines or making meal plans -- where humans reason routinely. This paper establishes the benefits of long CoT for general-purpose chat capabilities and introduces RL with Model-rewarded Thinking (RLMT)1, which pushes RLVR beyond verifiable domains. Using diverse real-world prompts, RLMT requires LMs to generate long CoT reasoning before responding, and optimizes them with online RL against a preference-based reward model used in RLHF. Across 40 training runs on Llama-3.1-8B and Qwen-2.5-7B (both base and instruct) and multiple optimization algorithms (DPO, PPO, and GRPO), RLMT consistently outperforms standard RLHF pipelines. This includes substantial gains of 3-7 points on three chat benchmarks (AlpacaEval2, WildBench, and ArenaHardV2), along with 1-3 point improvements on other tasks like creative writing and general knowledge. RLMT can also be applied directly to base models without an SFT stage, akin to DeepSeek-R1-Zero. Remarkably, with only 7K prompts, Llama-3.1-8B base trained with our RLMT recipe outperforms Llama-3.1-8B-Instruct post-trained with a complex multi-staged pipeline with 25M+ examples. We close with qualitative and quantitative analyses of how trained models plan their responses. Our results rethink the post-training pipeline and call upon future work to understand and employ thinking more broadly.
Computation and Language (cs.CL)
arXiv:2509.20357 [cs.CL]
(or arXiv:2509.20357v2 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2509.20357 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 324 words
Related coverage
Presentation: Choosing Your AI Copilot: Maximizing Developer Productivity
InfoQ · Jun 3, 2026
Links #5: 2026/07
LessWrong · Jul 31, 2026
Presentation: From OTEL to SLMs: Distilling Frontier Model Behaviour from Production Telemetry
InfoQ · Jul 17, 2026
[AINews] The Future of Latent Space
Latent Space · Sep 25, 2026