paper_id string | title string | published_date string | arxiv_categories list | legal_license string | license_url string | commercial_use_allowed bool | primary_author string | total_authors_count int64 | authors list | affiliations list | arxiv_url string | code_audit_status string | primary_repo_url string | repo_urls list | github_stars int64 | github_forks int64 | github_last_commit string | github_license string | importance_score float64 | abstract string | title_vector_384d list | abstract_vector_384d list | target_industry string | execution_mode string | agent_architectures list | model_parameter_scale string | min_inference_vram_fp16_gb float64 | min_inference_vram_quantized_gb float64 | recommended_gpu_class string | serving_frameworks_compatible list | tested_benchmarks list | extracted_benchmark_scores list | reproduction_recipe string | core_problem_addressed string | key_technical_innovation string | key_quantitative_result string | trend_velocity_tier string | papers_last_30_days int64 | trl_score int64 | readiness_tier string | trl_justification string | audited_at string |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
2608.09290v2 | OpenCodeReview: Determinism over Non-Determinism for Cost-Effective Agent-Based Code Review | 2026-08-10T08:43:16Z | [
"cs.SE"
] | ArXiv Standard | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | true | Zhengfeng Li | 11 | [
"Zhengfeng Li",
"Lei Zhang",
"Xianwei Wu",
"Zhengqi Zhuang",
"Yingjie Xu",
"Boge Wang",
"Shaofei Zhu",
"Chuan Wang",
"Peng Zhao",
"Xinyu Zheng",
"Guoping Rong"
] | [
"Alibaba Group"
] | http://arxiv.org/abs/2608.09290v2 | VERIFIED_LIVE | https://github.com/alibaba/open-code-review | [
"https://github.com/alibaba/open-code-review"
] | 20,650 | 0 | Unknown | Unspecified | 157.52 | LLM-based code review agents promise scalable, always-on review, yet current systems suffer from two intertwined weaknesses: (1) non-determinism--unbounded tool use makes review outcomes unstable, and (2) context locality--the reviewer's access remains bounded to the diff, capping discoverable issue depth. Both give ri... | [
-0.02571200020611286,
-0.020723000168800354,
-0.1330839991569519,
0.06809200346469879,
0.038899000734090805,
-0.009310999885201454,
-0.0071669998578727245,
-0.049157001078128815,
0.010979999788105488,
0.024511000141501427,
0.0024800000246614218,
-0.0178849995136261,
0.020569000393152237,
-... | [
-0.09384600073099136,
-0.043609000742435455,
-0.10136300325393677,
0.07402999699115753,
0.07184500247240067,
-0.06302899867296219,
-0.012698999606072903,
-0.015407999977469444,
0.01586800068616867,
0.019740000367164612,
-0.023541999980807304,
-0.027303999289870262,
0.03233100101351738,
-0.... | Autonomous AI Agents, Multi-Agent Swarms & Tool Use | Cloud API Orchestration (Frontier LLM) | [
"Self-Reflective & ReAct Loop",
"Tool-Calling & Model Context Protocol (MCP)",
"Autonomous Software Engineering Agent"
] | Cloud API (GPT-4o / Claude 3.5 Sonnet / Gemini 1.5) | 0 | 0 | Cloud API Endpoint (No local GPU required) | [
"LangChain / CrewAI Orchestrator",
"LiteLLM API Proxy",
"AutoGen Multi-Agent Runtime",
"Model Context Protocol (MCP) Host"
] | [
"Tool-Calling API Execution & ToolBench Protocols"
] | [
"Qualitative Trajectory Verification & Task Success Metrics"
] | git clone https://github.com/alibaba/open-code-review && cd open-code-review && (pip install -e . || pip install -r requirements.txt) | LLM-based code review agents promise scalable, always-on review, yet current systems suffer from two intertwined weaknesses: (1) non-determinism--unbounded tool use makes review outcomes unstable, and (2) context locality--the reviewer's access remains bounded to the diff, capping discoverable issue depth. | To address these, we introduce OpenCodeReview, built on deterministic engineering for uncertain agents: rather than granting maximal freedom, we inject determinism at three deliberate pipeline points. | On AACR-Bench (200 real-world PRs, 10 languages, 1,505 expert-verified comments), OpenCodeReview outperforms mainstream coding agents (e.g., Claude Code and Codex) across six LLM backends, achieving up to 2.17x higher SEM-F1 (25.10% vs. 11.57%) while consuming 5-15x fewer tokens. | Explosive (>50/mo) | 787 | 10 | Production-Ready (TRL 7-8) | Score 10/10. Live GitHub Repo (+2); High Stars (20650★) (+2); Production Agent Architecture (+2) | 2026-08-17T15:54:50.596909 |
2607.24223v2 | A New Role for Relevance: Guiding Corpus Interaction in Agentic Search | 2026-07-27T09:56:52Z | [
"cs.CL"
] | ArXiv Standard | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | true | Jiangnan Li | 5 | [
"Jiangnan Li",
"Yuqing Li",
"Mo Yu",
"Jinchao Zhang",
"Jie Zhou"
] | [] | http://arxiv.org/abs/2607.24223v2 | VERIFIED_LIVE | https://github.com/NVIDIA/NeMo-Retriever | [
"https://github.com/NVIDIA/NeMo-Retriever",
"https://github.com/LeqsNaN/RARG",
"https://github.com/texttron/RISE"
] | 2,964 | 0 | Unknown | Unspecified | 135.75 | Relevance is a query-dependent estimate of whether a document or excerpt contains useful evidence. Existing retrieval agents use relevance to select top-$k$ content, but document relevance alone cannot localize, compose, or verify the evidence required by complex questions. Direct Corpus Interaction (DCI) enables such ... | [
0.004738999996334314,
0.01921899989247322,
-0.014348000288009644,
-0.000307999987853691,
0.10312200337648392,
0.06095699965953827,
0.06055000051856041,
0.023504000157117844,
0.041572000831365585,
-0.0596579983830452,
0.02212199941277504,
-0.0034260000102221966,
0.006269000004976988,
0.0108... | [
-0.0456399992108345,
-0.01916399970650673,
0.018124999478459358,
0.02973400056362152,
0.11796200275421143,
0.04255300015211105,
0.04941900074481964,
0.07709699869155884,
0.03372599929571152,
-0.06714499741792679,
-0.016157999634742737,
0.024949999526143074,
-0.040754999965429306,
-0.030065... | Autonomous AI Agents, Multi-Agent Swarms & Tool Use | Agentic Protocol & Workflow Engine | [
"Autonomous Agent Core"
] | Model-Agnostic / Architecture Framework | 0 | 0 | CPU Server / Model-Agnostic Host | [
"LangGraph State Machine",
"CrewAI Flow",
"Model Context Protocol (MCP)",
"Custom Python Agent Loop"
] | [
"Empirical Agent Trajectory & Task Completion Analysis"
] | [
"Qualitative Trajectory Verification & Task Success Metrics"
] | git clone https://github.com/NVIDIA/NeMo-Retriever && cd NeMo-Retriever && (pip install -e . || pip install -r requirements.txt) | Relevance is a query-dependent estimate of whether a document or excerpt contains useful evidence. | Existing retrieval agents use relevance to select top-$k$ content, but document relevance alone cannot localize, compose, or verify the evidence required by complex questions. | These results demonstrate that relevance-aware interaction enables faster and more reliable search convergence. | Explosive (>50/mo) | 787 | 8 | Production-Ready (TRL 7-8) | Score 8/10. Live GitHub Repo (+2); High Stars (2964★) (+2) | 2026-08-17T15:55:21.482359 |
2608.04588v1 | EASy: Towards Efficient LLM-Based Agentic System | 2026-08-05T08:50:36Z | [
"cs.CL",
"cs.AI"
] | ArXiv Standard | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | true | Junnan Liu | 4 | [
"Junnan Liu",
"Linhao Luo",
"Thuy-Trang Vu",
"Gholamreza Haffari"
] | [] | http://arxiv.org/abs/2608.04588v1 | VERIFIED_LIVE | https://github.com/huggingface/Math-Verify | [
"https://github.com/bytedance/SandboxFusion",
"https://github.com/huggingface/Math-Verify"
] | 1,182 | 0 | Unknown | Unspecified | 126.22 | Agentic systems have emerged as a promising paradigm for solving complex tasks by coordinating specialized LLM-based agents. However, most existing systems primarily optimize task success while giving limited consideration to execution efficiency under practical constraints such as executor capability and computational... | [
0.024219999089837074,
-0.011091000400483608,
-0.05008300021290779,
-0.07040199637413025,
0.036107998341321945,
-0.02977300062775612,
0.020377999171614647,
-0.03369700163602829,
-0.025032000616192818,
0.06297799944877625,
-0.013830999843776226,
-0.014635000377893448,
0.0949999988079071,
-0.... | [
0.00035799999022856355,
-0.019864000380039215,
-0.023795999586582184,
0.01963699981570244,
-0.000022000000171829015,
-0.03938499838113785,
-0.07124499976634979,
-0.02465200051665306,
-0.023865999653935432,
0.021340999752283096,
-0.06476400047540665,
-0.004745000042021275,
0.01133500039577484... | Autonomous AI Agents, Multi-Agent Swarms & Tool Use | Agentic Protocol & Workflow Engine | [
"Autonomous Agent Core"
] | Model-Agnostic / Architecture Framework | 0 | 0 | CPU Server / Model-Agnostic Host | [
"LangGraph State Machine",
"CrewAI Flow",
"Model Context Protocol (MCP)",
"Custom Python Agent Loop"
] | [
"Interactive Sandbox & Environment Simulation"
] | [
"Qualitative Trajectory Verification & Task Success Metrics"
] | git clone https://github.com/huggingface/Math-Verify && cd Math-Verify && (pip install -e . || pip install -r requirements.txt) | Agentic systems have emerged as a promising paradigm for solving complex tasks by coordinating specialized LLM-based agents. | However, most existing systems primarily optimize task success while giving limited consideration to execution efficiency under practical constraints such as executor capability and computational cost. | Extensive experiments on mathematical reasoning, embodied decision-making, and deep research benchmarks show that EASy consistently achieves stronger performance-efficiency trade-offs than strong agentic baselines. | Explosive (>50/mo) | 787 | 9 | Production-Ready (TRL 7-8) | Score 9/10. Live GitHub Repo (+2); High Stars (1182★) (+2); Agent Toolkit / Benchmark (+1) | 2026-08-17T15:54:59.342420 |
2607.13027v1 | PalmClaw: A Native On-Device Agent Framework for Mobile Phones | 2026-07-14T17:58:57Z | [
"cs.CL",
"cs.AI"
] | ArXiv Standard | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | true | Hongru Cai | 4 | [
"Hongru Cai",
"Yongqi Li",
"Ran Wei",
"Wenjie Li"
] | [] | http://arxiv.org/abs/2607.13027v1 | VERIFIED_LIVE | https://github.com/ModalityDance/PalmClaw | [
"https://github.com/ModalityDance/PalmClaw"
] | 1,153 | 0 | Unknown | Unspecified | 124.86 | Large Language Model (LLM) agents have moved beyond generating responses to executing multi-step tasks by calling tools, observing the results, and iteratively deciding the next action. Most agent systems run on desktops or servers, which support tool use and task automation. Mobile devices are also important agent env... | [
-0.0683009997010231,
-0.010227999649941921,
0.010901999659836292,
-0.0682860016822815,
0.06847900152206421,
-0.04039999842643738,
0.013782000169157982,
-0.02024099975824356,
-0.043616000562906265,
-0.02155500091612339,
0.06169800087809563,
-0.03999200090765953,
0.05561399832367897,
-0.0430... | [
0.010677999816834927,
-0.03520900011062622,
0.01716800034046173,
-0.06990300118923187,
0.010281000286340714,
-0.04421300068497658,
-0.000699999975040555,
0.006401000078767538,
0.03173099830746651,
-0.011353000067174435,
0.04189600050449371,
-0.012651000171899796,
0.0651950016617775,
0.0143... | Autonomous AI Agents, Multi-Agent Swarms & Tool Use | Agentic Protocol & Workflow Engine | [
"Tool-Calling & Model Context Protocol (MCP)"
] | Model-Agnostic / Architecture Framework | 0 | 0 | CPU Server / Model-Agnostic Host | [
"LangGraph State Machine",
"CrewAI Flow",
"Model Context Protocol (MCP)",
"Custom Python Agent Loop"
] | [
"Interactive Sandbox & Environment Simulation"
] | [
"Qualitative Trajectory Verification & Task Success Metrics"
] | git clone https://github.com/ModalityDance/PalmClaw && cd PalmClaw && (pip install -e . || pip install -r requirements.txt) | Large Language Model (LLM) agents have moved beyond generating responses to executing multi-step tasks by calling tools, observing the results, and iteratively deciding the next action. | Most agent systems run on desktops or servers, which support tool use and task automation. | Experiments show an 11.5\% relative improvement in task success and a 94.9\% reduction in completion time over the strongest baseline, with lower setup burden and traces illustrating how execution boundaries are applied. | Explosive (>50/mo) | 787 | 10 | Production-Ready (TRL 7-8) | Score 10/10. Live GitHub Repo (+2); High Stars (1153★) (+2); Production Agent Architecture (+2) | 2026-08-17T15:55:31.390161 |
2607.13125v2 | Boogu-Image-0.1: Boosting Open Agentic Multimodal Generation via Understanding under a Minimal Budget | 2026-07-14T17:52:05Z | [
"cs.CV",
"cs.AI"
] | ArXiv Standard | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | true | Guoxuan Chen | 33 | [
"Guoxuan Chen",
"Chufeng Xiao",
"Haoran Yang",
"Siyue Xie",
"Binxiao Huang",
"Ming Zhang",
"Cheuk Him Chau",
"Xinyu Fu",
"Yingzhao Lian",
"Tom S. Y. Li",
"Jintao Lin",
"Bowen Dong",
"Zian Qian",
"Yuhao Liu",
"Yuxuan Hu",
"Weikang Shi",
"Bin Zou",
"Bowen Zheng",
"Haoxuan Che",
"... | [] | http://arxiv.org/abs/2607.13125v2 | VERIFIED_LIVE | https://github.com/Boogu-Project/Boogu-Image | [
"https://github.com/Boogu-Project/Boogu-Image"
] | 952 | 0 | Unknown | Unspecified | 122.78 | We introduce Boogu-Image-0.1, an open-source unified multimodal understanding and generation model family, comprising Base, Turbo, Edit, and Edit-Turbo variants. It delivers competitive performance in high-quality text-to-image generation, fast inference, instruction-based editing, and bilingual (Chinese-English) text ... | [
0.014453000389039516,
-0.03585999831557274,
-0.03346500173211098,
0.0060339998453855515,
0.14772799611091614,
-0.05170400068163872,
-0.005805999971926212,
-0.08183299750089645,
0.026115000247955322,
-0.010200999677181244,
0.0015660000499337912,
-0.11851699650287628,
0.05055300146341324,
0.... | [
0.004131999798119068,
-0.08541300147771835,
0.009030000306665897,
0.01536600012332201,
0.09495499730110168,
-0.07065500319004059,
-0.1171019971370697,
-0.03931400179862976,
0.02595599927008152,
-0.002672000089660287,
0.014706999994814396,
-0.07248300313949585,
0.04865400120615959,
0.001664... | Autonomous AI Agents, Multi-Agent Swarms & Tool Use | Local Open-Weights LLM | [
"Autonomous Agent Core"
] | 7B - 8B (Standard Agentic Model - Llama-3/Qwen) | 16 | 5.5 | Consumer GPU (RTX 4090 / 24GB) | [
"vLLM (High Throughput)",
"Ollama / llama.cpp (Local)",
"SGLang (Fast Agent Scheduling)",
"TensorRT-LLM"
] | [
"Multi-Agent Debate Consensus & Human Evaluation"
] | [
"Qualitative Trajectory Verification & Task Success Metrics"
] | git clone https://github.com/Boogu-Project/Boogu-Image && cd Boogu-Image && (pip install -e . || pip install -r requirements.txt) | We introduce Boogu-Image-0.1, an open-source unified multimodal understanding and generation model family, comprising Base, Turbo, Edit, and Edit-Turbo variants. | Closed-source multimodal systems like Nano-Banana-Pro and GPT-Image-2 achieve strong performance through system-level integration rather than a single model, yet their internal practices remain largely undisclosed. | Comprehensive evaluations show that Boogu-Image-0.1 consistently matches or surpasses other open-source models across standard benchmarks, and achieves results approaching leading closed-source systems. | Explosive (>50/mo) | 787 | 9 | Production-Ready (TRL 7-8) | Score 9/10. Live GitHub Repo (+2); High Stars (952★) (+2); Agent Toolkit / Benchmark (+1) | 2026-08-17T15:55:31.518159 |
2607.25333v2 | Specula: Scaling formal specifications for autonomous model checking of system code | 2026-07-28T06:33:15Z | [
"cs.SE",
"cs.AI",
"cs.DC",
"cs.OS"
] | ArXiv Standard | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | true | Qian Cheng | 9 | [
"Qian Cheng",
"Saad Mohammad Rafid Pial",
"Ruize Tang",
"Yiming Su",
"Emilie Ma",
"Finn Hackett",
"Ivan Beschastnikh",
"Yu Huang",
"Tianyin Xu"
] | [] | http://arxiv.org/abs/2607.25333v2 | VERIFIED_LIVE | https://github.com/specula-org/Specula | [
"https://github.com/specula-org/Specula"
] | 386 | 0 | Unknown | Unspecified | 113.69 | Specula is a push-button agentic system that generates high-quality formal specifications for large, complex system code and uses the specifications for highly effective model checking and bug finding. Specula employs large language model (LLM) based coding agents to autonomously develop TLA+ specifications, including ... | [
-0.01989700086414814,
-0.0021959999576210976,
-0.06337299942970276,
-0.007172999903559685,
0.04044799879193306,
-0.09495499730110168,
0.014295999892055988,
0.0492670014500618,
-0.03859800100326538,
0.03224800154566765,
-0.0023980000987648964,
-0.06527400016784668,
0.0172520000487566,
-0.02... | [
-0.0005450000171549618,
-0.01624000072479248,
-0.07513300329446793,
-0.040428999811410904,
-0.023876000195741653,
-0.08758199959993362,
-0.016172999516129494,
0.06500700116157532,
-0.03408199921250343,
0.014929999597370625,
-0.020191999152302742,
-0.0442579984664917,
0.021026000380516052,
... | Autonomous AI Agents, Multi-Agent Swarms & Tool Use | Agentic Protocol & Workflow Engine | [
"Autonomous Agent Core"
] | Model-Agnostic / Architecture Framework | 0 | 0 | CPU Server / Model-Agnostic Host | [
"LangGraph State Machine",
"CrewAI Flow",
"Model Context Protocol (MCP)",
"Custom Python Agent Loop"
] | [
"Empirical Agent Trajectory & Task Completion Analysis"
] | [
"Qualitative Trajectory Verification & Task Success Metrics"
] | git clone https://github.com/specula-org/Specula && cd Specula && (pip install -e . || pip install -r requirements.txt) | Specula is a push-button agentic system that generates high-quality formal specifications for large, complex system code and uses the specifications for highly effective model checking and bug finding. | Specula employs large language model (LLM) based coding agents to autonomously develop TLA+ specifications, including invariants that describe correctness properties of the target system and formal models that describe the system implementation with the right level of abstractions. | Meanwhile, Specula addresses limitations of LLM-driven techniques like reward hacking and hallucinations through self-evolving loops that iteratively improve specification quality by enabling the agents to deepen their understanding of system code and its behaviors. | Explosive (>50/mo) | 787 | 7 | Prototype (TRL 4-6) | Score 7/10. Live GitHub Repo (+2); Active Stars (386★) (+1) | 2026-08-17T15:55:15.462396 |
2607.14777v1 | SEED: Self-Evolving On-Policy Distillation for Agentic Reinforcement Learning | 2026-07-16T09:57:18Z | [
"cs.CL"
] | ArXiv Standard | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | true | Jinyang Wu | 11 | [
"Jinyang Wu",
"Shuo Yang",
"Zhengxi Lu",
"Fan Zhang",
"Yuhao Shen",
"Lang Feng",
"Haoran Luo",
"Zheng Lian",
"Shuai Zhang",
"Zhengqi Wen",
"Jianhua Tao"
] | [] | http://arxiv.org/abs/2607.14777v1 | VERIFIED_LIVE | https://github.com/jinyangwu/SEED | [
"https://github.com/jinyangwu/SEED"
] | 239 | 0 | Unknown | Unspecified | 107.91 | Large language models are increasingly trained as interactive agents for long-horizon tasks involving multi-turn interaction, tool use, and environment feedback. Outcome-based reinforcement learning (RL) provides a practical optimization paradigm, but its sparse trajectory-level rewards offer limited guidance on interm... | [
-0.05433500185608864,
-0.10269799828529358,
0.08572299778461456,
0.005427999887615442,
0.04126900061964989,
-0.028970999643206596,
0.021900000050663948,
-0.06773299723863602,
0.0002789999998640269,
-0.010831999592483044,
-0.052035000175237656,
0.028046999126672745,
-0.009867999702692032,
-... | [
-0.02865700051188469,
-0.04813599959015846,
0.04676799848675728,
0.04719800129532814,
0.07546599954366684,
0.02941500023007393,
0.02286599949002266,
-0.009998000226914883,
0.01618799939751625,
-0.024862999096512794,
-0.05829299986362457,
0.005547999870032072,
0.00024900000425986946,
0.0549... | Autonomous AI Agents, Multi-Agent Swarms & Tool Use | Local Open-Weights LLM | [
"Tool-Calling & Model Context Protocol (MCP)"
] | 7B - 8B (Standard Agentic Model - Llama-3/Qwen) | 16 | 5.5 | Consumer GPU (RTX 4090 / 24GB) | [
"vLLM (High Throughput)",
"Ollama / llama.cpp (Local)",
"SGLang (Fast Agent Scheduling)",
"TensorRT-LLM"
] | [
"Interactive Sandbox & Environment Simulation"
] | [
"Qualitative Trajectory Verification & Task Success Metrics"
] | git clone https://github.com/jinyangwu/SEED && cd SEED && (pip install -e . || pip install -r requirements.txt) | Large language models are increasingly trained as interactive agents for long-horizon tasks involving multi-turn interaction, tool use, and environment feedback. | We propose SEED (SElf-Evolving On-Policy Distillation), a self-evolving framework that converts completed on-policy trajectories into training-time hindsight skills and distills their behavioral effect back into the policy model. | Extensive experiments on text-based and vision-based agentic tasks show that SEED consistently improves performance and sample efficiency, exhibiting robust generalization to unseen scenarios. | Explosive (>50/mo) | 787 | 9 | Production-Ready (TRL 7-8) | Score 9/10. Live GitHub Repo (+2); Active Stars (239★) (+1); Production Agent Architecture (+2) | 2026-08-17T15:55:30.167831 |
2607.21268v1 | pAI-Econ-claude: A Gated Human-in-the-Loop Multi-Agent Architecture for AI-Assisted Economic Theory Development | 2026-07-23T12:40:47Z | [
"cs.MA",
"cs.AI",
"econ.GN"
] | ArXiv Standard | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | true | Chen Zhu | 3 | [
"Chen Zhu",
"Xiaolu Wang",
"Weilong Zhang"
] | [] | http://arxiv.org/abs/2607.21268v1 | VERIFIED_LIVE | https://github.com/maxwell2732/pAI-Econ-claude | [
"https://github.com/maxwell2732/pAI-Econ-claude"
] | 152 | 0 | Unknown | Unspecified | 103.37 | In many social-science research tasks, such as economics, LLM-based agents must produce outputs for which no cheap, task-complete, machine-readable correctness signal exists. This creates a distinctive reliability problem for multi-agent systems: how should generation, critique, coordination, and human judgment be orga... | [
-0.012961000204086304,
-0.032926999032497406,
-0.10021200031042099,
-0.030734000727534294,
-0.0039900001138448715,
-0.0021569998934865,
0.003126000054180622,
0.009279999881982803,
0.04769200086593628,
-0.02156900055706501,
-0.04174499958753586,
0.0015829999465495348,
-0.04222499951720238,
... | [
-0.020541999489068985,
-0.01897899992763996,
-0.10350800305604935,
0.004344000015407801,
0.053001001477241516,
-0.017734000459313393,
-0.005204999819397926,
0.010707000270485878,
0.029603999108076096,
0.010146000422537327,
-0.08469600230455399,
0.003693999955430627,
0.06282100081443787,
0.... | Autonomous AI Agents, Multi-Agent Swarms & Tool Use | Cloud API Orchestration (Frontier LLM) | [
"Autonomous Agent Core"
] | Cloud API (GPT-4o / Claude 3.5 Sonnet / Gemini 1.5) | 0 | 0 | Cloud API Endpoint (No local GPU required) | [
"LangChain / CrewAI Orchestrator",
"LiteLLM API Proxy",
"AutoGen Multi-Agent Runtime",
"Model Context Protocol (MCP) Host"
] | [
"Empirical Agent Trajectory & Task Completion Analysis"
] | [
"Qualitative Trajectory Verification & Task Success Metrics"
] | git clone https://github.com/maxwell2732/pAI-Econ-claude && cd pAI-Econ-claude && (pip install -e . || pip install -r requirements.txt) | In many social-science research tasks, such as economics, LLM-based agents must produce outputs for which no cheap, task-complete, machine-readable correctness signal exists. | This creates a distinctive reliability problem for multi-agent systems: how should generation, critique, coordination, and human judgment be organized when no component can certify the final result? | The results support a bounded claim: gated oversight improves the auditability of AI-assisted economic theory without substituting for formal verification, and the allocation of irreversible human judgment is a more informative design variable than pure agent autonomy. | Explosive (>50/mo) | 787 | 7 | Prototype (TRL 4-6) | Score 7/10. Live GitHub Repo (+2); Active Stars (152★) (+1) | 2026-08-17T15:55:21.130018 |
2607.22529v1 | Skill Self-Play: Pushing the Frontier of LLM Capability with Co-Evolving Skills | 2026-07-24T17:59:22Z | [
"cs.CL"
] | ArXiv Standard | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | true | Siyuan Huang | 13 | [
"Siyuan Huang",
"Pengyu Cheng",
"Haotian Liu",
"Tao Chen",
"Yihao Liu",
"Jingwei Ni",
"Shijie Zhou",
"Ziyi Yang",
"Gangwei Jiang",
"Mengyu Zhou",
"Yu Cheng",
"Xiaoxi Jiang",
"Guanjun Jiang"
] | [] | http://arxiv.org/abs/2607.22529v1 | VERIFIED_LIVE | https://github.com/Qwen-Applications/skill-self-play | [
"https://github.com/Qwen-Applications/skill-self-play"
] | 125 | 0 | Unknown | Unspecified | 101.31 | LLM training is shifting from manual design and annotation to interaction-driven self-evolution. However, existing self-evolutionary methods face a fundamental dilemma between task diversity and verification reliability: environment-bound methods obtain precise feedback but confine learning to narrow domains, while ope... | [
0.047189000993967056,
-0.09129799902439117,
-0.006252000108361244,
-0.06248699873685837,
-0.008933999575674534,
0.03140399977564812,
0.0417729988694191,
-0.009461999870836735,
-0.07534299790859222,
0.005901999771595001,
-0.03817300125956535,
-0.023374000564217567,
0.002093000104650855,
0.0... | [
-0.03980899974703789,
-0.06847099959850311,
-0.013109000399708748,
0.006022000219672918,
-0.0026990000624209642,
0.00634300010278821,
0.030667999759316444,
-0.0030539999715983868,
-0.0241870004683733,
-0.005229999776929617,
-0.08022599667310715,
-0.07526899874210358,
-0.001572999986819923,
... | Autonomous AI Agents, Multi-Agent Swarms & Tool Use | Local Open-Weights LLM | [
"Tool-Calling & Model Context Protocol (MCP)"
] | 7B - 8B (Standard Agentic Model - Llama-3/Qwen) | 16 | 5.5 | Consumer GPU (RTX 4090 / 24GB) | [
"vLLM (High Throughput)",
"Ollama / llama.cpp (Local)",
"SGLang (Fast Agent Scheduling)",
"TensorRT-LLM"
] | [
"Interactive Sandbox & Environment Simulation"
] | [
"Qualitative Trajectory Verification & Task Success Metrics"
] | git clone https://github.com/Qwen-Applications/skill-self-play && cd skill-self-play && (pip install -e . || pip install -r requirements.txt) | LLM training is shifting from manual design and annotation to interaction-driven self-evolution. | Leveraging this insight, we introduce Skill Self-Play (Skill-SP), a co-evolutionary framework comprising a proposer, a solver, and a dynamic skill controller. | Empirical evaluations on tool-use and reasoning benchmarks demonstrate that Skill-SP, serving as a robust evolution engine, consistently pushes the performance ceiling of competent backbones while catalyzing striking turnarounds for initially misaligned models. | Explosive (>50/mo) | 787 | 8 | Production-Ready (TRL 7-8) | Score 8/10. Live GitHub Repo (+2); Active Stars (125★) (+1); Agent Toolkit / Benchmark (+1) | 2026-08-17T15:55:18.580563 |
2607.28430v1 | AgentRadio: Passive Awareness for Long-Horizon Multi-Agent Collaboration | 2026-07-30T16:07:32Z | [
"cs.MA"
] | ArXiv Standard | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | true | Xinxing Ren | 7 | [
"Xinxing Ren",
"Qianbo Zang",
"Ziyan Wang",
"Caelum Forder",
"Suman Deb",
"Peter Carroll",
"Zekun Guo"
] | [] | http://arxiv.org/abs/2607.28430v1 | VERIFIED_LIVE | https://github.com/Coral-Protocol/AgentRadio | [
"https://github.com/Coral-Protocol/AgentRadio"
] | 109 | 0 | Unknown | Unspecified | 100.13 | "Understanding large codebases is a long-horizon task for Large Language Model (LLM) agents: answeri(...TRUNCATED) | [0.06829699873924255,-0.05094600096344948,-0.0852459967136383,-0.01577799953520298,0.027253000065684(...TRUNCATED) | [-0.010195000097155571,-0.044725000858306885,-0.043258000165224075,-0.027775000780820847,0.024875000(...TRUNCATED) | Autonomous AI Agents, Multi-Agent Swarms & Tool Use | Cloud API Orchestration (Frontier LLM) | [
"Autonomous Agent Core"
] | Cloud API (GPT-4o / Claude 3.5 Sonnet / Gemini 1.5) | 0 | 0 | Cloud API Endpoint (No local GPU required) | ["LangChain / CrewAI Orchestrator","LiteLLM API Proxy","AutoGen Multi-Agent Runtime","Model Context (...TRUNCATED) | [
"Empirical Agent Trajectory & Task Completion Analysis"
] | [
"Qualitative Trajectory Verification & Task Success Metrics"
] | "git clone https://github.com/Coral-Protocol/AgentRadio && cd AgentRadio && (pip install -e . || pip(...TRUNCATED) | "Understanding large codebases is a long-horizon task for Large Language Model (LLM) agents: answeri(...TRUNCATED) | "On SWE-Atlas QnA, a benchmark of long-horizon questions over production repositories, a single Clau(...TRUNCATED) | "Under a five-phase protocol of division of labor and negotiation, four agents organized by AgentRad(...TRUNCATED) | Explosive (>50/mo) | 787 | 8 | Production-Ready (TRL 7-8) | Score 8/10. Live GitHub Repo (+2); Active Stars (109★) (+1); Agent Toolkit / Benchmark (+1) | 2026-08-17T15:55:10.591946 |
End of preview. Expand in Data Studio
🤖 Autonomous AI Agents & Multi-Agent Swarms Dataset (2023–2026)
Sample dataset of 30 audit-verified research papers covering Autonomous AI Agents, Multi-Agent Swarms, Tool Calling, and Model Context Protocols (MCP) with 384d PyTorch embeddings.
🛒 Full 1,000 Paper B2B Dataset Available on Gumroad
Get the complete 3-year dataset (1,000 papers + VRAM & Execution Modes + SQLite/CSV/Parquet + Quickstart Script) on Gumroad: 👉 Get Full 1,000 Dataset on Gumroad ($19 / $39 / $89)
- Downloads last month
- 48