OpenAI o1 System Card

By Ahmed El-Kishky. arXiv.org, 2024

1,946 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/985298bd0b6f4c84ab34bb15cfed0b092f43b83a

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

Citations in the corpus

Process Reinforcement through Implicit Rewards (2025, positive): “OpenAI o1 (Jaech et al., 2024) first showed the tremendous potential of large-scale RL for reasoning LLMs, and recent works have verified the scaling effect of RL with outcome rewards (DeepSeek-AI et al., 2025; Team et al., 2025).”
Cognitive Behaviors that Enable Self-Improving Reasoners, or, Four Habits of Highly Effective STaRs (2025, positive): “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,…”
Video-R1: Reinforcing Video Reasoning in MLLMs (2025, neutral): “Recent advancements in rule-based Reinforcement Learning (RL) [11] have significantly enhanced the reasoning capabilities of Large Language Models (LLMs) [10, 6].”
VLM-R1: A Stable and Generalizable R1-style Large Vision-Language Model (2025, positive): “The introduction of OpenAI o1 [20] demonstrated that re-inforcement learning (RL), which enables Large Language Models (LLMs) to directly learn from feedback on their outputs, can significantly enhance their reasoning capabilities.”
The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity (2025, neutral): “Large Language Models (LLMs) have recently evolved to include specialized variants explicitly designed for reasoning tasks—Large Reasoning Models (LRMs) such as OpenAI’s o1/o3 [1, 2], DeepSeek-R1 [3], Claude 3.7 Sonnet Thinking [4], and Gemini Thinking [5].”
GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models (2025, neutral): “While state-of-the-art proprietary models like OpenAI’s o1/o3 [18] and Anthropic’s Claude Sonnet 4 have demonstrated groundbreaking performance in specific ARC domains (e.g., mathematical reasoning or code fixing [20]), a single, powerful open-source model that excels across all three areas has…”
Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model? (2025, neutral): “We also let OpenAI-o1 (Jaech et al., 2024) • RLVR narrows the reasoning boundary.”
GLM-4.5V and GLM-4.1V-Thinking: Towards Versatile Multimodal Reasoning with Scalable Reinforcement Learning (2025, positive): “Recently, numerous studies have shown that long-form reasoning [60] and scalable reinforcement learning [42] can significantly enhance the ability of large language models (LLMs) to solve complex problems [23; 17].”
WebThinker: Empowering Large Reasoning Models with Deep Research Capability (2025, positive): “Models like OpenAI-o1 [21], Qwen-QwQ [50], and DeepSeek-R1 [5] demonstrate explicit chain-of-thought (CoT) reasoning [56], resembling human system-2 thinking in tasks such as math and programming [29].”
Vision-R1: Incentivizing Reasoning Capability in Multimodal Large Language Models (2025, positive): “As researchers have discovered that enabling LLMs to simulate human-like thought processes and perform stepwise reasoning can significantly enhance their performance on reasoning tasks [1], extensive work has been dedicated to exploring LLM reasoning methods.”
VL-Rethinker: Incentivizing Self-Reflection of Vision-Language Models with Reinforcement Learning (2025, neutral): “The release of GPT-o1 [Jaech et al., 2024] and DeepSeek-R1 [Guo et al., 2025] has sparked renewed interest in incentivizing reasoning capabilities in LLMs via reinforcement learning (RL).”
Flow-GRPO: Training Flow Matching Models via Online RL (2025, positive): “At the same time, online reinforcement learning (RL) [9] has proven highly effective in enhancing the reasoning capabilities of large language models (LLMs) [10, 11].”
Kimi k1.5: Scaling Reinforcement Learning with LLMs (2025, positive): “…intelligence, reinforcement learning (RL) has emerged as a pivotal training methodology for large language models (LLMs)(Ouyang et al. 2022)(Jaech et al. 2024), drawing inspiration from its success in mastering complex games like Go, StarCraft II, and Dota 2 through systems such as…”
Emerging Properties in Unified Multimodal Pretraining (2025, positive): “Inspired by recent models like O1 [33] and DeepSeek-R1 [26], we leverage long-context Chain-of-Thoughts data for multimodal understanding.”
Dream 7B: Diffusion Large Language Models (2025, neutral): “…an additional dimension for inference-time scaling (Snell et al., 2024; Muennighoff et al., 2025; Geiping et al., 2025), which works alongside existing techniques such as chain-of-thought reasoning used in large language models like OpenAI o1 (Jaech et al., 2024) and DeepSeek R1 (Guo et al., 2025).”
SimpleRL-Zoo: Investigating and Taming Zero Reinforcement Learning for Open Base Models in the Wild (2025, neutral): “Large reasoning models, including OpenAI-o1 (Jaech et al., 2024), DeepSeek-R1 (DeepSeek-AI et al., 2025a), and Kimi-k1.”
SFT Memorizes, RL Generalizes: A Comparative Study of Foundation Model Post-training (2025, positive): “Notably, recent findings (Jaech et al., 2024; DeepSeekAI et al., 2025) reveal “scaling laws” for inference-time compute, highlighting significant performance gains with increased computational resources.”
Search-R1: Training LLMs to Reason and Leverage Search Engines with Reinforcement Learning (2025, positive): “Notably, models like OpenAI-o1 (Jaech et al., 2024) and DeepSeek-R1 (Guo et al., 2025) have leveraged RL techniques ( e.g. , PPO (Schulman et al., 2017) and GRPO (Shao et al., 2024)) to improve logical inference and problem-solving skills by learning from experience and feedback.”
π0.5: a Vision-Language-Action Model with Open-World Generalization (2025, neutral): “…same exact model for both high-level and low-level inference, in a recipe that more closely resembles chain-of-thought [82] or test-time compute [39] methods, though unlike embodied chain-of-thought methods [88, 46, 61], the high-level inference process still runs at a lower frequency than…”