BigBang: Pursuing Open-Ended Intelligence through Self-Evolving Synthesis of Verifiable Frontier Tasks

The BigBang Team

BigBang-v1 is a 35B-A3B agentic model trained with a self-evolving generator-critic pipeline for verifiable frontier tasks. It leads comparable 35B models across eight benchmarks, scoring 76.5% on BrowseComp, 50.3% on HLE, 54.2% on SWE-Bench Pro, 53.6% on PaperBench (Code-Dev), and 46.2% on FrontierScience Research, with 16K+ Hugging Face downloads.

Paper / Code / Model / 机器之心

AI4AI at Scale: Building Open-Weight Deep Search Agents

XYZ Team

XYZ-Aquila-pro achieves strong results across six benchmarks: 84.8% on BrowseComp, 85.1% on BrowseComp-ZH, 92.5% on DeepSearchQA, 53.7% on LiveBrowseComp, 53.3% on Humanity’s Last Exam, and 81.2% on WideSearch. The OpenSeeker data also provided important support for these capability gains.

Paper / Harness_code / Training_code / 机器之心

OpenSeeker-v2: Pushing the Limits of Search Agents with Informative and High-Difficulty Trajectories

Yuwen Du^*, Rui Ye^*, Shuo Tang, Keduan Huang, Xinyu Zhu, Yuzhu Cai, Siheng Chen

A compact SFT release that expands OpenSeeker with more informative and higher-difficulty trajectories, showing that only 10k samples can still push same-scale pure ReAct models to SOTA performance.

ArXiv / Code / Model

OpenSeeker: Democratizing Frontier Search Agents by Fully Open-Sourcing Training Data

Yuwen Du^*, Rui Ye^*, Shuo Tang, Xinyu Zhu, Yijun Lu, Yuzhu Cai, Siheng Chen

We fill a long-standing gap in frontier search by fully open-sourcing the training data and model, making strong search agents more reproducible and accessible.

ArXiv / Code / Training Data / Model / 机器之心

PaSaMaster: Towards Self-Evolving Agentic Literature Retrieval

Yuwen Du^*, Tian Jin^*, Jing Kang, Xianghe Pang, Jingyi Chai, Tingjia Miao, Fenyi Liu, WenHao Wang, Sikai Yao, Yuzhi Zhang, Siheng Chen

A recursive self-evolving agentic literature retrieval system that iteratively analyzes intent, retrieves verified papers, and ranks them with evidence-grounded relevance scores.

ArXiv / Code / Benchmark

SciMaster: Towards General-Purpose Scientific AI Agents, Part I. X-Master as Foundation: Can We Lead on Humanity’s Last Exam?

Jingyi Chai^*, Shuo Tang^*, Rui Ye^*, Yuwen Du^*, Xinyu Zhu, Mengcheng Zhou, Yanfeng Wang, Weinan E, Yuzhi Zhang, Linfeng Zhang, Siheng Chen

A general-purpose scientific AI agent built upon our tool-augmented reasoning agent X-Master, designed to flexibly interact with external tools during scientific reasoning.

ArXiv / Code

RoCo-Sim: Enhancing Roadside Collaborative Perception through Foreground Simulation

Yuwen Du^*, Anning Hu^*, Zichen Chao, Yifan Lu, Junhao Ge, Genjia Liu, Weitao Wu, Lanjun Wang, Siheng Chen

A foreground simulation framework that improves roadside collaborative perception by generating more realistic training scenes.

ArXiv / Code