Suncong Zheng
—
LLM researcher spanning model building and software-engineering evaluation — MoE reports and testing-grade benchmarks
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-15) 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—
No underlying output data — total publication count, venue mix, or per-year paper counts — is available in the current snapshot, so this axis is unscored. It would measure how much and where this researcher publishes.
Impact—
Citation counts, h-index-type indicators, and downstream uptake are absent from the supplied data, so impact cannot be scored. This axis would measure how widely the two cataloged papers and other works are cited or reused.
Focus0/100
Focus scores 0 because the field distribution is split exactly evenly — 50% software engineering (LLM evaluation) and 50% machine learning (large language models) — across the cataloged record. Rather than a missing signal, this reads as genuine dual-track work: model-side contributions (Hunyuan-A13B) on one side and evaluation-side contributions (WebCraftBench) on the other, with no single dominant specialization yet visible in the metrics.
Collaboration85/100
The collaboration score of 85 rests on a broad footprint: 71 distinct coauthors in the underlying bibliometric record, with a recurring core of twelve frequent collaborators including the Tencent Hunyuan Team, Chenxu Liu, Zilu Zou, and Ying Zhou. Both 2026 catalog papers are large-team, middle-author contributions — typical of industrial LLM technical reports and multi-author benchmark efforts — suggesting this researcher operates comfortably inside sizeable, distributed teams rather than leading small groups.
Activity—
Per-year publication activity is unavailable; the snapshot holds only two 2026 papers, which cannot establish cadence. This axis would measure publishing rhythm and recent momentum across multiple years.
Reproducibility38/100
Reproducibility averages 38 across the only two papers assessed (reproPapers: 2). Hunyuan-A13B rates medium — broad results tables but no variance, hardware specifications, or ablations — while WebCraftBench rates low despite strong validation (85.3% human agreement), because its dataset and code are closed-source. The mean is a thin, early-stage signal from a two-paper sample, not a settled track record.
Research summary
Suncong Zheng works at the intersection of large language models and software engineering, and the Evolve catalog captures both directions of that interface. The 2026 snapshot contains two arXiv papers, in each case with a middle author position rather than first or corresponding authorship. The first, the Hunyuan-A13B Technical Report, documents Tencent's 80B-total/13B-active mixture-of-experts model with chain-of-thought reasoning, GRPO-based reinforcement learning, and a 256K-token context; Evolve's verdict is that the reported near-frontier scores are promising but statistically unverified. The second, WebCraftBench, evaluates LLM web-application generation from a software-testing perspective, combining code coverage, automated GUI testing, and LLM-as-a-judge, with 85.3% human agreement, while its dataset and code remain closed-source. Bibliometrically the record splits evenly between machine learning (large language models) and software engineering (LLM evaluation), and the collaboration footprint is wide: 71 distinct coauthors, including the Tencent Hunyuan Team. Reproducibility readiness averages 38 over the two assessed papers, a modest early-stage signal. Output, impact, and activity metrics are not yet available, making this an early-stage profile name-resolved from public bibliographic data.
Collaboration pattern
The collaboration pattern is large-team and industry-flavored: 71 distinct coauthors overall, with a stable core of twelve frequent collaborators — Chenxu Liu, Zilu Zou, Peizhong Gao, Jiawen Tao, Zhexin Zhang, Guang Chen, Haowei Lin, Ying Zhou, Tianyi Bai, Dolly Deng, Maxm Pan, and the Tencent Hunyuan Team. Both catalog papers place this researcher in a middle author position, consistent with specialist contributions inside sizeable model-building and benchmark teams rather than sole or corresponding leadership.
Research keywords
Highlights
Co-authorship on Tencent's Hunyuan-A13B MoE technical report
Evaluating LLM web-app generation from a software-testing angle
Timeline
- 2026Co-authors the Hunyuan-A13B Technical Report (arXiv), a middle-position contribution to Tencent's 80B-total/13B-active MoE model covering chain-of-thought, GRPO reinforcement learning, and 256K context.
- 2026Co-authors WebCraftBench (arXiv), a coverage-guided benchmark for LLM web-application generation validated at 85.3% human agreement.
- 2026The two 2026 outputs split evenly across fields — 50% machine learning (large language models), 50% software engineering (LLM evaluation) — producing the lowest focus score.
- 2026The bibliometric record for 2026 lists 71 distinct coauthors, with the Tencent Hunyuan Team among twelve frequent collaborators.
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.
- Identity is name-resolved from public bibliographic data; no ORCID, institution, or country was supplied, and the Chinese-character form of the name is unconfirmed.
- Output, impact, and activity dimensions are unscored because publication counts, citation data, and per-year activity are absent from the snapshot.
- The reproducibility mean (38) averages only 2 assessed papers, both 2026 arXiv preprints, so it is an early-stage signal.
- Coverage is limited to papers reviewed in the Evolve catalog; other works under this name are out of scope, and radar scores derive from OpenAlex-based bibliometrics only.