Your daily digest
Research papers, made simple to understand
The AI and machine learning papers people are reading most, each one rewritten in plain words: the problem, the idea, what was found and why it matters.
- AI Agents
- Multimodal & Vision
Sage S2.1 Flash: A 320B-parameter multimodal model excelling in scientific reasoning and agentic tasks
Sage S2.1 Flash integrates scientific perception, multimodal reasoning, and agentic tool use in a 320B-parameter mixture-of-experts model, achieving top ranks on 33 benchmarks across scientific workflows, coding, tool orchestration, and visual interaction.
Paper: Sage S2.1 Flash Technical Report · Kun Cai, Haodong Chen, Guangran Cheng and 203 others
85 upvotes on Hugging Face6 min readRead the article
- Language Models & Architectures
- AI Agents
K2-Horizon fleet offers open-source LLMs with sparse attention and efficient training
K2-Horizon presents six open language models from 0.9B to 375B parameters, introducing Mixture-of-Value Attention and softplus gating to improve reasoning and efficiency, with full training data, code, and checkpoints released for reproducibility.
Paper: K2-Horizon: A Frontier Fleet of Open Source Language Models · K2 Team, Zhengzhong Liu, Haonan Li and 92 others
72 upvotes on Hugging Face7 min readRead the article
- AI Agents
- Reinforcement Learning & Alignment
AllocFlow generates joint portfolio allocations from raw market data with improved returns
AllocFlow uses a frozen market encoder and a flow Transformer to jointly generate continuous portfolio weights, trained first by imitation of a strategy and then refined by group-relative reinforcement learning, doubling returns and improving risk metrics on crypto futures.
Paper: AllocFlow: End-to-End Portfolio Trading Action Generation from Raw Market Data · Annan Wang, Yuchen Xie, Yu Cen and 5 others
67 upvotes on Hugging Face7 min readRead the article
Browse earlier papers
18 days of papers in November 2026
- WeVisForensics jointly trains an investigator agent and evolving forensic inspectors for AI video detectionand 2 more papers
- Jeveration generates images and dense predictions in one fast forward passand 2 more papers
- Xing 4 advances large language models on Ascend chips with efficient architecture and trainingand 2 more papers
- Sage S2.1 Flash: A 320B-parameter multimodal model excelling in scientific reasoning and agentic tasksYou are reading this day
- Agentic robotics uses foundation models plus harnesses for versatile physical task controland 2 more papers
- Direct Preference Optimization fine-tunes language models from human preferences without reinforcement learningand 2 more papers
- Layer normalization stabilizes and speeds up training, especially in recurrent networksand 2 more papers
- SO-Arena benchmarks System One models' decision capabilities and reliability across domainsand 2 more papers
- Structured textbook-based supervision improves mathematical reasoning in large language modelsand 2 more papers
- Tree-Ising optimizes sparse bit-level edits to fix reasoning loops in language modelsand 2 more papers
- Unified tactile representation improves dexterous robot manipulation across sensor typesand 2 more papers
- Gaussian Spacing Normalization improves feed-forward 3D Gaussian Splatting generalizationand 2 more papers
- VFlash enables fast, accurate visual plan decoding by jointly predicting entire primitivesand 2 more papers
- TurnWise improves streaming video RL by assigning credit at each decision turnand 1 more paper
- Routing specialized AI models boosts efficiency and performance across tasksand 2 more papers
- Unified taxonomy and literature system organize agentic computer graphics researchand 2 more papers
- FrameForge adapts video models for diverse conditional image generation with input-adaptive temporal relationsand 2 more papers
- Diffusion models generate high-quality images by learning to reverse noise additionand 2 more papers