PaperCodex

Flow-GRPO: Boost Text-to-Image Accuracy with Online RL—Without Sacrificing Quality or Diversity 1720

If you’ve ever struggled with diffusion models failing to follow detailed prompts—like “a golden retriever sitting to the left of…

12/19/2025Controllable Diffusion Models, Reinforcement Learning For Generative Models, Text-to-Image Generation

Memento: Build Smarter LLM Agents That Learn from Experience—Without Fine-Tuning 2060

In today’s fast-paced AI landscape, teams building intelligent agents face a persistent dilemma: how to make large language models (LLMs)…

12/19/2025Agent-based Reasoning, Continual Learning, Memory-Augmented LLMs

Matrix-Game: Controllable, Real-Time Game World Generation with Pixel-Perfect Action Responsiveness 1768

Matrix-Game is an open-source interactive world foundation model developed by Skywork AI, specifically designed for real-time, controllable generation of game…

12/19/2025Action-conditioned Simulation, Controllable Video Generation, Interactive World Modeling

FlowTok: Unified Text-to-Image and Image-to-Text Generation with Compact 1D Tokens 1082

FlowTok reimagines cross-modal generation by collapsing the traditionally complex boundary between text and images into a streamlined, efficient process. Unlike…

12/19/2025Image-to-text Generation, Multimodal Representation Learning, Text-to-Image Generation

Decompile-Bench: The First Million-Scale Real-World Benchmark for Training and Evaluating LLM-Powered Binary Decompilers 6178

Decompiling machine code back into human-readable source remains one of the most challenging and valuable tasks in software engineering, cybersecurity,…

12/19/2025Binary Decompilation, Code Translation, Reverse Engineering

MiniMax-M1: The First Open-Weight Hybrid-Attention Model for Long-Context Reasoning and Efficient AI Agents 3001

MiniMax-M1 is a breakthrough in open large language models: it’s the world’s first open-weight, large-scale hybrid-attention reasoning model. Designed for…

12/19/2025Agentic Tool Use, Long-context Reasoning, Software Engineering Agents

Hunyuan3D 2.1: Open-Source, High-Fidelity 3D Generation from Images with Production-Ready PBR Materials 2498

Creating high-quality 3D assets has long been a bottleneck in industries like gaming, virtual reality, industrial design, and digital content…

12/19/20253D Generation, Image-to-3D, PBR Material Synthesis

Reasoning Gym: Train and Evaluate Reasoning Models with Infinite, Verifiable Reinforcement Learning Environments 1265

If you’re building or evaluating reasoning-capable AI systems—especially large language models (LLMs)—you’ve likely hit a wall with static benchmarks. Traditional…

12/19/2025Procedural Task Generation, Reasoning, Reinforcement Learning

SageAttention3: 5x Faster LLM Inference on Blackwell GPUs with Plug-and-Play FP4 Attention and First-Ever 8-Bit Training Support 2814

Attention mechanisms lie at the heart of modern large language models (LLMs) and multimodal architectures—but their quadratic computational complexity remains…

12/19/2025Efficient Training, Large Language Model Inference, Multimodal Generation

Meta-World+: A Reproducible, Standardized Benchmark for Multi-Task and Meta Reinforcement Learning in Robotic Control 1659

Evaluating reinforcement learning (RL) agents—especially those designed for multi-task or meta-learning scenarios—requires benchmarks that are consistent, well-documented, and technically accessible.…

12/19/2025Meta-reinforcement Learning, Multi-task Reinforcement Learning, Robotic Manipulation