DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning

By Adam Suma, Sam Dauncey. arXiv.org, 2025

4,594 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/bf76cb3bd7a54c60138f9523ffb462a486c4807b

Citations in the corpus

• Search-R1: Training LLMs to Reason and Leverage Search Engines with Reinforcement Learning (2025, unknown): “Despite demonstrating remarkable reasoning (Guo et al., 2025) and coding (Guo et al., 2024) capabilities, LLMs (Zhao et al., 2023; Team, 2024; Achiam et al., 2023) often lack domain-specific knowledge (Peng et al., 2023; Li et al., 2023) and are prone to hallucinations (Zhang et al., 2023).”
• Cognitive Behaviors that Enable Self-Improving Reasoners, or, Four Habits of Highly Effective STaRs (2025, unknown): “A notable breakthrough came with Deepseek-R1 (Guo et al., 2025; Jaech et al., 2024), which demonstrated that even a simplified version of PPO (Schulman et al., 2017) called GRPO (Shao et al., 2024) (2025) have begun to unpack these results, revealing that supervised fine-tuning with long, structured”
• Kimi K2: Open Agentic Intelligence (2025, unknown): “Budget Control It has been widely observed that RL often results in a substantial increase in the length of model-generated responses [36, 20].”
• Large Language Diffusion Models (2025, unknown): “Finally, a systematic investigation into post-training for LLaDA (e.g., O1-like systems [99, 100]) is needed to further unlock the potential of diffusion language models.”
• Visual-RFT: Visual Reinforcement Fine-Tuning (2025, unknown): “Although the implementation details of o1 are not publicly available, recent open-source studies like DeepSeek R1 [4] reveal one promising direction in reproducing o1 is Verifiable Re-wards [4, 14, 39]: the reward score in reinforcement learning is directly determined by pre-defined rules, rather…”
• R1-Onevision: Advancing Generalized Multimodal Reasoning Through Cross-Modal Formalization (2025, unknown): “We use DeepSeek-R1 [8] to generate reasoning processes on LLaVA-OneVision [16] and collect these data into the R1-Onevision dataset.”
• Jailbreak and Guard Aligned Language Models With Only Few In-Context Demonstrations (2023, unknown): “5 [76], DeepSeek-R1 1 [77]).”
• Open-Reasoner-Zero: An Open Source Approach to Scaling Up Reinforcement Learning on the Base Model (2025, unknown): “Scaling RL on Reasoning-Enhanced Models Another important line of work [2, 8, 19, 20] endeavors to scale up RL training on reasoning-enhanced Models.”
• Vision-R1: Incentivizing Reasoning Capability in Multimodal Large Language Models (2025, unknown): “Recently, DeepSeek-R1 [2] has successfully applied Re-inforcement Learning (RL) to induce the self-emergence of complex cognitive reasoning ability in LLMs.”
• MARS: Multi-hop Adaptive Retrieval and SPARQL Generation for KGQA (2026, unknown): “Notable examples such as Gemini [7], Claude [1], and DeepSeek-R1 [10] achieve state-of-the-art performance on a wide variety of reasoning tasks [22].”
• An Empirical Study of Knowledge Graph-Enhanced RAG for Information Security Compliance (2026, unknown): “Three language models were chosen for further evaluation: deepseek-r1:8b [59], qwen2.”
• BRIDGE: Multimodal-to-Text Retrieval via Reinforcement-Learned Query Alignment (2026, unknown): “Reinforcement learning has recently emerged as a powerful paradigm for query optimization, motivated by the success of RLHF in LLM alignment [31] and reasoning [11].”
• Benchmarking Small Language Models and Small Reasoning Language Models on System Log Severity Classification (2026, unknown): “5B and Phi-4-Mini-Reasoning are distilled from larger reasoning models and explicitly optimized to produce long chain-of-thought outputs [88, 89].”
• ReasonIR: Training Retrievers for Reasoning Tasks (2025, unknown): “Weller et al. (2025) train a reranking model for reasoning-intensive queries by obtaining reasoning traces from DeepSeek-R1 (Guo et al., 2025) on MS-Marco queries and documents, and using these traces to train a distilled Qwen2.”
• Mieb: Massive Image Embedding Benchmark (2025, unknown): “We hypothesize that future models that better incorporate reasoning capabilities and test-time scaling techniques [20, 24, 37, 43, 62] may achieve better results on compositionality tasks.”
• LLM-confidence reranker: A training-free approach for enhancing retrieval-augmented generation systems (2026, unknown): “Leveraging their advanced capabilities in language understanding and generation, LLMs have significantly enhanced task efficiency and user experience across a wide range of applications (DeepSeek-AI et al., 2025).”