Contributor on LLM agent benchmark environments and edge–cloud multimodal inference routing
2
papers here
1
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-17) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.
Research profile
Output—
Not yet scored: the catalog holds too few indexed outputs to compute a reliable output measure. This axis would track publication volume and pace over time; once more of Zijian Wang's papers are ingested, Evolve can score it.
Impact—
Not yet scored: citation and influence data are unavailable in the current snapshot. This axis would measure how widely the work, such as the RecreationWorld benchmark, is adopted, cited, or built upon by other groups.
Focus100/100
Scores 100/100 for focus. In the current snapshot the field distribution contains a single entry — machine learning / LLM agents (1 paper) — so every field-classified output sits in one area. The reviewed work on hybrid computer-use agents and, separately, multimodal LLM inference routing both sit within the LLM systems space, consistent with a tightly concentrated agenda. Note this rests on one classified paper, so the perfect score should be read as concentration of classified evidence rather than demonstrated breadth.
Collaboration74/100
Scores 74/100. The snapshot records 39 distinct coauthors, with a recurring core including Shuai Bai, Jiayong Deng, Sicheng Fan, Yikun Fu, Chang Gao, Mianqiu Huang, Yizhen Jiang, Yuheng Jing, Dehui Kong, Keliang Li and Ning Li. Both cataloged papers are middle-author positions on multi-author arXiv papers, indicating work inside larger team efforts rather than solo projects; the coauthor count relative to few indexed papers suggests broad but early-stage collaborative reach.
Activity—
Not yet scored: publication-timing data is too sparse to measure activity. This axis would track output cadence — recent papers, posting rhythm, and sustained engagement across the full record; with only two 2026 arXiv entries visible, no reliable signal exists.
Reproducibility100/100
Scores 100/100, based on 1 paper with a reproducibility assessment. The assessed paper, RecreationWorld, is rated high for repro readiness: Evolve's verdict notes a five-platform recreation benchmark with a verifiable-reward SFT pipeline and rigorous evaluation design, while flagging that reported transfer gains lack statistical controls. With only one paper assessed, the perfect score should be treated as a strong single signal rather than a stable track record.
Research summary
Zijian Wang appears in the Evolve catalog as a contributor to two 2026 arXiv papers at the intersection of large language model agents and efficient multimodal inference, holding middle-author positions on both. The more fully reviewed of the two, RecreationWorld, introduces a five-platform recreation benchmark with scalable, verifiable environments for hybrid computer-use agents that combine GUI operation with coding capability, together with a supervised fine-tuning pipeline built around verifiable rewards. Evolve's assessment rates its evaluation design as rigorous and its reproducibility readiness as high, while cautioning that reported transfer gains lack statistical controls. The second paper, Pro-Router, proposes token-aware progressive model routing with adaptive edge–cloud collaboration to reduce the cost of multimodal LLM inference. Bibliometric signals in the snapshot are concentrated: the single field-classified paper sits in machine learning / LLM agents, giving a perfect focus score, while 39 distinct coauthors — including recurring collaborators such as Shuai Bai, Jiayong Deng, Chang Gao, Mianqiu Huang, Dehui Kong, Keliang Li and Ning Li — support a collaboration score of 74. Output, impact, and activity axes are not yet scored because citation and timing data are insufficient. This profile is name-resolved from public bibliographic data; no institution, country, or ORCID is attached.
Collaboration pattern
The snapshot lists 39 distinct coauthors, with a stable recurring cluster: Shuai Bai, Jiayong Deng, Sicheng Fan, Yikun Fu, Chang Gao, Mianqiu Huang, Yizhen Jiang, Yuheng Jing, Dehui Kong, Keliang Li, and Ning Li. Both cataloged papers are middle-author contributions to multi-author arXiv teams, consistent with large-group work in agent benchmarking and systems-efficiency research rather than solo or two-author projects.
Research keywords
Highlights
RecreationWorld: five-platform verifiable environments for hybrid computer-use agents
High reproducibility readiness for the reviewed agent benchmark
Pro-Router: token-aware progressive routing with edge–cloud collaboration
A second 2026 arXiv paper proposes adaptive edge–cloud model routing to cut multimodal LLM inference cost; it is not yet reviewed by Evolve, so no verdict or repro rating exists.
Timeline
- 2026RecreationWorld posted on arXiv: a five-platform recreation benchmark with scalable, verifiable environments and an SFT pipeline for hybrid GUI/coding agents.
- 2026Evolve review rates RecreationWorld's evaluation design as rigorous and its reproducibility readiness as high, while flagging missing statistical controls on transfer gains.
- 2026Pro-Router posted on arXiv, proposing token-aware progressive routing with adaptive edge–cloud collaboration for efficient multimodal LLM inference.
Papers in the Evolve catalog
Frequent co-authors
About this profile
This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.
- Bibliometric measures (39 coauthors, field distribution, radar scores) derive from Evolve's public bibliographic snapshot; citation counts and h-index are not part of the current data.
- Only two 2026 arXiv papers are in the catalog; Pro-Router lacks a field label, review verdict, keywords, and reproducibility rating.
- No institution, country, or ORCID is available; Zijian Wang is a common name, so this profile is name-resolved from public bibliographic data and may merge or split identities.
- The reproducibility score rests on a single assessed paper (reproPapers = 1) and may shift as more work is reviewed.