Sage S2.1 Flash Technical Report

We introduce Sage S1.5 Flash, a 320-billion-parameter general-purpose mixture-of-experts model designed to reason over scientific evidence and act in interactive environments. Sage Sci-Perception uses dedicated encoders for molecular graphs and biological sequences and represents spectra as ordered patches of measurements. Modality-specific projectors map these representations into the input space of a shared backbone, which processes them alongside text and images. To develop the model's agent capabilities, we construct verifiable environments for tool use, search, and coding, as well as specialized workflows involving machine learning experiments and scientific software. Task completion is checked against execution results, changes in environment state, or submitted artifacts. We evaluate Sage S1.5 Flash on 33 benchmarks across seven domains. In the primary comparison against five baselines, it ranks first on OfficeQA Pro (72.2), MLE-bench Lite (90.3 HumanRank), Harvey LAB, and SpreadsheetBench~2, and second on OSWorld-Verified (85.2), WebArena-Verified (67.6), and DeepResearch Bench~II (54.2). These results show particular strengths in using documentary evidence, producing verifiable workbooks, and carrying out machine learning tasks, alongside competitive performance in desktop and browser workflows that require the model to act on visual feedback.

Authors

Publication Details

Journal
arXiv (Cornell University)
Published
2026-11-16
DOI
https://doi.org/10.48550/arxiv.2611.21072
Primary Topic
Multimodal Machine Learning Applications
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Sage S2.1 Flash Technical Report

Tianhao Liang, 蔡奇燁, Yuhang Zang, Qiushi He et al.
arXiv (Cornell University)
Multimodal Machine Learning Applications
preprint

Sage S2.1 Flash Technical Report

Tianhao Liang, 蔡奇燁, Yuhang Zang, Qiushi He, Guanting Dong, Junlin Shang, Chenyang Gu, Xinmeng Che, Jiabei Chen, Xiaoyu Xing, Ningsheng Ma, Wenqing Jing, Guangran Cheng, Wenjie Fu, Jiadong Chen, Dingbo Yuan, Shuhao Xing, Qiming Ge, Ruojun Zhou, Alexander Lam, Binze Hu, Huayu Sha, Haijun Lv, Qisheng Su, Zhongbo Tian, Huicheng Jiang, Feiyang Hao, Ran Tao, Jiazhen Jiang, Jiahang Lin, Tao Gui, Kun Cai, Zijian Hu, Peng Li, Jianmin Qian, Ding Han, Lixin Gu, Qi Li, Yicheng Zou, Wanghan Xu, Tong Liu, Shuo Shen, Honglin Guo, Hongji Chen, Guoqiang Wang, Jianing Geng, Junhao Huang, Rui Wang, Qingxiu Cheng, Enfu Huang, Xinyu Yang, Haoran Zhao, Jiajun Sun, Bo Li, Jiaxiang Liu, Yichen Dong, Xinchen Xie, Yujia Chen, Yifei Cao, Jiayi Chen, Zhecan Wang, Yuyang Hu, Changhao Jiang, Xu Dai, Zijian Hu, Junzhang Liu, Xin Hong, Qilong Li, Jiahao Chen, Jiajie Jin, Yajie Yang, Yanjiang Liu, Zixian Huang, Yining Li, Zerun Ma, Shaofan Liu, Hongjie Zhang, Dongsheng Zhu, Penghao Zhao, Chengfeng Zhao, Yongkang Hu, Yanan Sun, Xinjie Chen, Hao Xiang, Xinyu Lu, Zhuoqun Li, Han Han, Yi Dong, Shufan Jiang, Xufeng He, Xin Deng, Shuainan Liu, Jie Hou, Jiahao Kuang, Hui Wang, Haodong Chen, Zhichao Sun, Yiding Wang, Minlong Peng, Jiaqiang Li
preprint en

Abstract

We introduce Sage S1.5 Flash, a 320-billion-parameter general-purpose mixture-of-experts model designed to reason over scientific evidence and act in interactive environments. Sage Sci-Perception uses dedicated encoders for molecular graphs and biological sequences and represents spectra as ordered patches of measurements. Modality-specific projectors map these representations into the input space of a shared backbone, which processes them alongside text and images. To develop the model's agent capabilities, we construct verifiable environments for tool use, search, and coding, as well as specialized workflows involving machine learning experiments and scientific software. Task completion is checked against execution results, changes in environment state, or submitted artifacts. We evaluate Sage S1.5 Flash on 33 benchmarks across seven domains. In the primary comparison against five baselines, it ranks first on OfficeQA Pro (72.2), MLE-bench Lite (90.3 HumanRank), Harvey LAB, and SpreadsheetBench~2, and second on OSWorld-Verified (85.2), WebArena-Verified (67.6), and DeepResearch Bench~II (54.2). These results show particular strengths in using documentary evidence, producing verifiable workbooks, and carrying out machine learning tasks, alongside competitive performance in desktop and browser workflows that require the model to act on visual feedback.

arXiv (Cornell University)
Multimodal Machine Learning Applications
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon