Lumine: An Open Recipe for Building Generalist Agents in 3D Open Worlds
On the map: LLM agents and tool use → Reinforcement learning for LLM agents
Abstract (arXiv)
We introduce Lumine, the first open recipe for developing generalist agents capable of completing hours-long complex missions in real time within challenging 3D open-world environments. Lumine adopts a human-like interaction paradigm that unifies perception, reasoning, and action in an end-to-end manner, powered by a vision-language model. It processes raw pixels at 5 Hz to produce precise 30 Hz keyboard-mouse actions and adaptively invokes reasoning only when necessary. Trained in Genshin Impact, Lumine successfully completes the entire five-hour Mondstadt main storyline on par with human-level efficiency and follows natural language instructions to perform a broad spectrum of tasks in both 3D open-world exploration and 2D GUI manipulation across collection, combat, puzzle-solving, and NPC interaction. In addition to its in-domain performance, Lumine demonstrates strong zero-shot cross-game generalization. Without any fine-tuning, it accomplishes 100-minute missions in Wuthering Waves and the full five-hour first chapter of Honkai: Star Rail. These promising results highlight Lumine's effectiveness across distinct worlds and interaction dynamics, marking a concrete step toward generalist agents in open-ended environments.
Key facts. Cited by 19 papers in the map (data through 7 Oct 2026). Placed in Reinforcement learning for LLM agents, one of 16 directions in LLM agents and tool use. First posted on arXiv 12 Nov 2025.
Builds on
0 of 0 references.
No references here yet.
Built on by
Citing papers per year: 2025: 2; 2026: 17.
19 papers.
Sort:
By area:
- 2025SIMA 2: A Generalist Embodied Agent for Virtual Worldscited by 26
- 2026NitroGen: An Open Foundation Model for Generalist Gaming Agentscited by 22
- 2026GameWorld: Towards Standardized and Verifiable Evaluation of Multimodal Game Agentscited by 22
- 2026Scaling Behavior Cloning Improves Causal Reasoning: An Open Model for Real-Time Video Game Playingcited by 10
- 2026GameVerse: Can Vision-Language Models Learn from Video-based Reflection?cited by 7
- 2026Research on World Models Is Not Merely Injecting World Knowledge into Specific Taskscited by 3
- 2026MapTab: A Diagnostic Benchmark for Long-Horizon Multi-Criteria Multimodal Reasoning on Heterogeneous Topological Graphscited by 3
- 2026Hera: Learning Long-Horizon Coordination for Device-Cloud Collaborative LLM Agentscited by 3
- 2025IPR-1: Interactive Physical Reasonercited by 2
- 2026MAIN-VLA: Modeling Abstraction of Intention and eNvironment for Vision-Language-Action Modelscited by 2
- 2026Odysseus: Scaling VLMs to 100+ Turn Decision-Making in Games via Reinforcement Learningcited by 2
- 2026Towards Generalist Game Players: An Investigation of Foundation Models in the Game Multiversecited by 2
7 more
- 2026SPIKE: An Adaptive Dual Controller Framework for Cost-Efficient Long-Horizon Game Agentscited by 2
- 2026SimpleTool: Parallel Decoding for Real-Time LLM Function Callingcited by 1
- 2026ActSWM: Action-Sensitive World Models for Long-Horizon Planning in Open-World Gamescited by 1
- 2026UrbanGround: From Local Perception to Spatial Agency in a Real-Scale Citycited by 1
- 2026GROW: Aligning GRPO with State-Action Modeling for Open-World VLM Agentscited by 0
- 2026GameWAM: A World Action Model for Video Gamescited by 0
- 2026GameHorizon Suite: Multi-Horizon Data and Evaluation in Gameplaycited by 0
- 2026GameHorizon Suite: Multi-Horizon Data and Evaluation in Gameplay21 Sep 2026
- 2026UrbanGround: From Local Perception to Spatial Agency in a Real-Scale City27 Aug 2026
- 2026GameWAM: A World Action Model for Video Games25 Aug 2026
- 2026ActSWM: Action-Sensitive World Models for Long-Horizon Planning in Open-World Games29 Jul 2026
- 2026Hera: Learning Long-Horizon Coordination for Device-Cloud Collaborative LLM Agents23 May 2026
- 2026GROW: Aligning GRPO with State-Action Modeling for Open-World VLM Agents18 May 2026
- 2026SPIKE: An Adaptive Dual Controller Framework for Cost-Efficient Long-Horizon Game Agents18 May 2026
- 2026Towards Generalist Game Players: An Investigation of Foundation Models in the Game Multiverse11 May 2026
- 2026Odysseus: Scaling VLMs to 100+ Turn Decision-Making in Games via Reinforcement Learning1 May 2026
- 2026GameWorld: Towards Standardized and Verifiable Evaluation of Multimodal Game Agents8 Apr 2026
- 2026GameVerse: Can Vision-Language Models Learn from Video-based Reflection?1 Mar 2026
- 2026MapTab: A Diagnostic Benchmark for Long-Horizon Multi-Criteria Multimodal Reasoning on Heterogeneous Topological Graphs20 Feb 2026
7 more
- 2026SimpleTool: Parallel Decoding for Real-Time LLM Function Calling4 Feb 2026
- 2026MAIN-VLA: Modeling Abstraction of Intention and eNvironment for Vision-Language-Action Models2 Feb 2026
- 2026Research on World Models Is Not Merely Injecting World Knowledge into Specific Tasks2 Feb 2026
- 2026Scaling Behavior Cloning Improves Causal Reasoning: An Open Model for Real-Time Video Game Playing8 Jan 2026
- 2026NitroGen: An Open Foundation Model for Generalist Gaming Agents4 Jan 2026
- 2025SIMA 2: A Generalist Embodied Agent for Virtual Worlds4 Dec 2025
- 2025IPR-1: Interactive Physical Reasoner19 Nov 2025
- 2025SIMA 2: A Generalist Embodied Agent for Virtual Worldscited by 26
- 2026NitroGen: An Open Foundation Model for Generalist Gaming Agentscited by 22
- 2026Hera: Learning Long-Horizon Coordination for Device-Cloud Collaborative LLM Agentscited by 3
- 2026SimpleTool: Parallel Decoding for Real-Time LLM Function Callingcited by 1
- 2026Hera: Learning Long-Horizon Coordination for Device-Cloud Collaborative LLM Agents23 May 2026
- 2026SimpleTool: Parallel Decoding for Real-Time LLM Function Calling4 Feb 2026
- 2026NitroGen: An Open Foundation Model for Generalist Gaming Agents4 Jan 2026
- 2025SIMA 2: A Generalist Embodied Agent for Virtual Worlds4 Dec 2025
- 2026GameWorld: Towards Standardized and Verifiable Evaluation of Multimodal Game Agentscited by 22
- 2026GameVerse: Can Vision-Language Models Learn from Video-based Reflection?cited by 7
- 2026Odysseus: Scaling VLMs to 100+ Turn Decision-Making in Games via Reinforcement Learningcited by 2
- 2026UrbanGround: From Local Perception to Spatial Agency in a Real-Scale Citycited by 1
- 2026UrbanGround: From Local Perception to Spatial Agency in a Real-Scale City27 Aug 2026
- 2026Odysseus: Scaling VLMs to 100+ Turn Decision-Making in Games via Reinforcement Learning1 May 2026
- 2026GameWorld: Towards Standardized and Verifiable Evaluation of Multimodal Game Agents8 Apr 2026
- 2026GameVerse: Can Vision-Language Models Learn from Video-based Reflection?1 Mar 2026
- 2026Research on World Models Is Not Merely Injecting World Knowledge into Specific Taskscited by 3
- 2025IPR-1: Interactive Physical Reasonercited by 2
- 2026ActSWM: Action-Sensitive World Models for Long-Horizon Planning in Open-World Gamescited by 1
- 2026GameWAM: A World Action Model for Video Gamescited by 0
- 2026GameWAM: A World Action Model for Video Games25 Aug 2026
- 2026ActSWM: Action-Sensitive World Models for Long-Horizon Planning in Open-World Games29 Jul 2026
- 2026Research on World Models Is Not Merely Injecting World Knowledge into Specific Tasks2 Feb 2026
- 2025IPR-1: Interactive Physical Reasoner19 Nov 2025
- 2026Scaling Behavior Cloning Improves Causal Reasoning: An Open Model for Real-Time Video Game Playingcited by 10
- 2026MapTab: A Diagnostic Benchmark for Long-Horizon Multi-Criteria Multimodal Reasoning on Heterogeneous Topological Graphscited by 3
- 2026GROW: Aligning GRPO with State-Action Modeling for Open-World VLM Agentscited by 0
- 2026GROW: Aligning GRPO with State-Action Modeling for Open-World VLM Agents18 May 2026
- 2026MapTab: A Diagnostic Benchmark for Long-Horizon Multi-Criteria Multimodal Reasoning on Heterogeneous Topological Graphs20 Feb 2026
- 2026Scaling Behavior Cloning Improves Causal Reasoning: An Open Model for Real-Time Video Game Playing8 Jan 2026
Similar, not cited
Close in wording, no citation either way
- 2026MineExplorer: Evaluating Open-World Exploration of MLLM Agents in Minecraft
- 2026OpenGameEval: Benchmarking Agentic Programming and Exploration in a Stateful Game Engine
- 2023JARVIS-1: Open-World Multi-task Agents with Memory-Augmented Multimodal Language Models
- 2024Craftium: Bridging Flexibility and Efficiency for Rich 3D Single- and Multi-Agent Environments
- 2025Game-TARS: Pretrained Foundation Models for Scalable Generalist Multimodal Game Agents
- 2026GS-Agent: Creating 4D Physical Worlds With Generative Simulation
- 2025OmniActor: A Generalist GUI and Embodied Agent for 2D&3D Worlds
- 2025OpenHA: A Series of Open-Source Hierarchical Agentic Models in Minecraft