Research Library
Library
Papers, patents & whitepapers, annotated
SkillRise: Agentic Reinforcement Learning for Cross-Task Skill Evolution
Zhiyuan Yao, Yuxin Chen, Zhengxi Lu et al. · arXiv · 2026
Notable for a test-time result — performance keeps improving as the sequence of related tasks grows longer, with no extra training. That is a different scaling axis from the usual ones.
Pass the Baton: Trajectory-Relayed On-Policy Distillation
Haolei Xu, Xiaowen Xu, Haiwen Hong et al. · arXiv · 2026
Training small models from larger ones is now a standard production step. This is a cheap, label-free fix for the failure mode that wastes most of that compute.
Shieldstral
Antonia Calvi, Avinash Sooriyarachchi, Giada Pistilli et al. · arXiv · 2026
Guardrails are usually the least documented part of a production stack. A small, open, policy-adaptable classifier is a practical way to run moderation without paying frontier-model prices on every call.
CLBench-V: Evaluating Multimodal Context Learning from Grounding to Knowledge Acquisition
Lai Wei, Chengqi Li, Jiapeng Li et al. · arXiv · 2026
A useful corrective if you plan to feed charts and scanned documents to a multimodal model. That headline number says the capability is far weaker than the demos imply.
CAST: Game Solvers as Turn-Level Teachers for LLM Agents
Yu Wang, Yi-Kai Zhang, Wentao Shi et al. · arXiv · 2026
A clean example of borrowing a signal from classical AI to fix a modern training problem. Wherever a solver already exists, dense per-step supervision turns out to be nearly free.
CodeNib: A Multi-View Data System for Serving Repository Context to Coding Agents
Zhongming Yu, Hengjia Yu, Boqin Yuan et al. · arXiv · 2026
The infrastructure story behind why some coding agents feel fast and cheap while others crawl. Most useful if you are building agent tooling over a codebase rather than consuming someone else's.
CoRT: Counterfactual Replay for Token-Level Rubric-Guided Policy Optimization
Bo-Wen Zhang, Junwei He, Wen Wang et al. · arXiv · 2026
Neat engineering: it extracts a token-level signal already latent in the setup instead of bolting on more machinery. Read it for how credit assignment actually works inside GRPO-style training.
DecoEvo: Score-Decoupled Co-Evolution of Solver and Rubric-Generator Skills in Text Space
Jiangwang Chen, Zixin Song, Junlin Liu et al. · arXiv · 2026
Relevant well past research: anyone running an LLM-as-judge pipeline is optimising against a fixed rubric and will meet the same ceiling. The decoupling argument is the part to take away.
From Proprietary to Open-Source: Bridging the Distribution Gap via Multi-Agent Protocol Distillation in Agentic Search
Junlin Liu, Jiangwang Chen, Zixin Song et al. · arXiv · 2026
A practical route for teams that want frontier-model behaviour from a model they can host themselves. The protocol-as-intermediate trick is the transferable idea; search is just where it gets demonstrated.
Kimi K3: Open Frontier Intelligence
Kimi Team, Tongtong Bai, Yifan Bai et al. · arXiv · 2026
The clearest current look at how an open-weight lab builds a frontier-scale model — sparsity, long context, and agentic RL documented in one place. Read it when you want the engineering behind the leaderboard numbers rather than the numbers themselves.
Keep It InMind: Benchmarking the Implicit-Association Blind Spot in Agent Memory
Ruizhe Li, Mingxuan Du, Benfeng Xu et al. · arXiv · 2026
The most striking result in this batch. If you are building on a vector-store memory layer, this is the failure mode you have not tested for — and larger embeddings do not close the gap.
A New Role for Relevance: Guiding Corpus Interaction in Agentic Search
Jiangnan Li, Yuqing Li, Mo Yu et al. · arXiv · 2026
Worth reading if you build RAG and keep hitting the ceiling of top-k retrieval. It reframes ranking as something that steers the search process rather than something that filters the final context window.
Mage-VL: An Efficient Codec-Native Streaming Multimodal Foundation Model
Senqiao Yang, Kaichen Zhang, Zhaoyang Jia et al. · arXiv · 2026
Interesting for what it borrows. Video compression solved 'which pixels carry information' decades ago, and applying that to tokenisation is a more elegant fix than simply sampling fewer frames.
The Physics of Multi-Turn Long-Horizon Planning: From Pre-training to Post-training via Single- and Multi-Teacher On-Policy Agentic Distillation
Tianyi Men, Zhuoran Jin, Kang Liu et al. · arXiv · 2026
Most agent papers report that a technique works; this one tries to explain why. The finding that suboptimal trajectories hurt more than they help is a direct warning about training on scraped agent traces.
Novel Claim or Déjà Vu? Rethinking "Contamination-Free'' Dynamic Evaluation for Multimodal Automated Fact-Checking
Haorui He, Xinwen Chen, Dacheng Wen et al. · arXiv · 2026
A caution worth internalising whenever you read a benchmark result. 'Published after the cutoff' is a weaker guarantee than it sounds, because reasoning over old knowledge can answer new questions.
JarvisHub: An Open Harness for Canvas-Native Multimodal Creative Agents
Yunlong Lin, Zixu Lin, Zhaohu Xing et al. · arXiv · 2026
A concrete answer to a problem every agent builder eventually hits: a chat transcript is a poor substitute for project state. The canvas-as-memory framing generalises well past creative tools.
StateAct: Program State, before Pixels, for Long-Horizon Computer-Use Agents
Yan Yang, Xiangru Jian, Ziyang Luo et al. · arXiv · 2026
One of the sharper recent arguments in agent design: pick the interface carrying the most information, not the one that most resembles how a person would do it. The cost figure is the part to remember.
Scaling Native Multimodal Pre-Training From Scratch
Haoyuan Wu, Aoqi Wu, Hai Wang et al. · arXiv · 2026
Scaling laws are how labs decide where compute goes before they spend it. This one says the text-only recipe does not carry over — the data mix changes the answer.
LAMAR: An Open Language-Aware Multilingual Alignment Reranker
Seongtae Hong, Youngjoon Jang, Jungseob Lee et al. · arXiv · 2026
A specific, easily-missed bug in multilingual RAG. If your users query in one language and your corpus spans several, your reranker probably has this problem right now.
Skill Self-Play: Pushing the Frontier of LLM Capability with Co-Evolving Skills
Siyuan Huang, Pengyu Cheng, Haotian Liu et al. · arXiv · 2026
Self-play is how systems surpassed human data in games, and this is a serious attempt to port that to general model capability. Read it alongside the other co-evolution work here to see the pattern forming.
Showing 1–20 of 30 sources · newest first
About Nybble™
The AI space moves fast.
Nybble™ is how you keep up — and stay sharp.
What happened. In two minutes.
The AI news cycle moves at a pace no one can keep up with. Snack distills what launched, what shipped, and what matters — every day, without the filler.
Go to SnackThe concepts behind the headlines.
News tells you what. Stack tells you why and how. From RAG architectures to agentic evals, these are the ideas that will shape what you build next.
Go to StackProve you actually get it.
Reading about LangChain is not the same as knowing it. Hack challenges you with production-grade questions, then shows you the references that make the answer stick.
Go to Hack