Can a coding agent take over a teammate’s work in progress?
Introducing SWE-CoWork, a benchmark for coding agents working as digital employees on long-horizon software engineering tasks.
Each assignment starts with a short handoff in a simulated workplace modeled on real-world engineering workflows. Agents search task boards and versioned docs, ask teammates for missing details, and revise their code when a reply changes what needs to be built.
Open questions and confirmed requirements can both be lost before the agent hands the work back. SWE-CoWork lets us trace them through the conversation, the patch, and the tests.
101 assignments across 101 real open-source repositories.
26 simulated teammates per workplace.
11 programming languages.
12 application domains.
17 workplace tools.
🌐 Website: tokenwave-ai.github.io/swe-c…
📄 Paper: arxiv.org/abs/2611.20218
💻 Code: github.com/TokenWave-AI/swe-…
🤗 Hugging Face Daily Papers: huggingface.co/papers/2611.2… (If you find it useful, we’d appreciate an upvote!)
To see where delivery breaks down, we analyzed a set of CoWork runs from Claude Opus 5.5.
We traced 979 failed checks for new behavior back to specific requirements. In 44% of those failures, the agent never received the requirement. In 56%, the agent had received it and the check still failed.
In one run on carapace-bin, the agent received all 25 requirements held by teammates. Still, 22 of 35 checks for new behavior failed. The trace showed incorrect CLI flags and a build failure the agent left unchecked.
If you’re building coding agents, pick a failed check and trace it back through the agent’s conversations, code changes, and tests.
1
1
119
A big thank you to everyone who helped build SWE-CoWork.
@LeiPing78377, Minghao Liu, @RidgerZhu , Xiaohui Hu, Xingqiao Li, Ao Zhang, @wenhaocha1 , Han Li, Tao Wang, Jingyuan Zhang, Enduo Zhao, Yunfeng Shi, Yiming Liang, Zetao Yang, Xingjian Wang, Yifan Yao, Xinkai Ma, @RichardYRLi , Tianhan Yang, Huan Zhou, Weiran Shi, @shen_82, @liujiaheng2 , Stephen Huang, Eli Cheung, and @ethanwhx.
1
153
[1/9]
What are the limits of AI agents in scientific research? Can they use professional software as effectively as scientists?
Meet OSWorld-Science: a benchmark and evaluation environment for computer-use agents, with tasks grounded in real-world scientific research needs—from pathology image analysis to physical simulation.
146 tasks.
7 scientific domains.
17 tools.
12 VLMs evaluated.
Tasks are developed through expert proposals and iterative human–AI co-design, with selection guided by scientific value and difficulty.
By combining GUI/CLI interaction, artifact-based evaluation, and trajectory analysis, we examine how agents carry out scientific workflows and where execution breaks down.
TokenWave is glad to be part of this research collaboration.
🌐 Project: discoailab.github.io/osworld…
📄 Paper: arxiv.org/abs/2609.39903
💻 Code: github.com/DiscoAILab/OSWorl…
📊 Data: huggingface.co/datasets/SciA…
[8/9]
Help expand OSWorld-Science:
• Contribute tasks grounded in real research needs across scientific, clinical, and engineering workflows.
• Bring cross-application and interdisciplinary tasks, including drug-discovery workflows.
• Debug and verify tasks and evaluators.
• Support testing with API access or sandbox resources.
Researchers and collaborators welcome.
Submit a task: forms.gle/dAK1xuLJCfGJG9sm7
1
101
[9/9]
Thank you to all our OSWorld-Science collaborators:
Dingyuan Dai, Heli Qi, Lei Liu, Yinxi Li, Baiding Chen, Zijun Dou;
Qingcheng Zeng (@SteveZeng7), Qi Kang, Oliver Kunyang Sun, Eric Wang, Bo Zhou, Haixin Wang, Yufan Du, Shi Bo, Ruihan Lin, Mengqi Yuan, Dunjie Lu, Steven Dillmann (@StevenDillmann), Yiming Shi, Tina Su, Amy Xin, Minghao Liu, Xi Wang, Xu Huang;
Ge Zhang (@GeZhang86038849), Pengyu Nie, Zhen Yang, Jie Tang (@jietang), Juanzi Li, Weihao Xuan, and Tianyu Liu (@meSuper8).
Full author list and affiliations: discoailab.github.io/osworld…
1
267
A closer look at YuE2: from prompt roulette to editable AI music.
YuE2 is a 3.58B-parameter open-weight music model that writes an editable melody-and-chord score before rendering a complete song with vocals and accompaniment.
We see YuE2 as a step toward a new paradigm for AI music: explicit musical planning, with a score that creators and agents can inspect and revise before it becomes audio.
The paper tests whether that planning helps. Using the same model checkpoint, 49.3% of expert judgments favored symbolic planning versus 34.6% without it; the rest were ties.
TokenWave supplied most of the project’s synthetic training data under license, as acknowledged in the paper. We’re glad to have contributed alongside HKUST, M-A-P, and the wider research team to this direction: making musical structure explicit, editable, and part of generation itself.
What does this make possible?
🎼 Edit the composition.
Change the melody or harmony, then render a new performance. The exact vocals and sound may vary between versions.
🎙️ Reimagine a song.
Use a transcribed melody to create a new interpretation in a different style, without cover-specific fine-tuning.
🤖 Collaborate with an agent.
An external language-model agent translates musical feedback into score, style, or lyric revisions. YuE2 generates the audio.
Under the hood, a unified AR–NAR Mixture-of-Transformers connects scores to audio through semantic tokens and flow-matched acoustic latents. MERT2 and SheetSage2 provide semantic and symbolic training supervision.
Demos and editable scores:
map-yue2.github.io/
Paper:
arxiv.org/abs/2609.33757
Code:
github.com/multimodal-art-pr…
1
3
170

