LLM agent evaluation researcher at Alibaba Group building multilingual and long-horizon benchmarks for autonomous agents
2
papers here
1
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
12
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—
Insufficient data: total publication volume and venue composition are not available in this snapshot beyond yearly counts. This axis would measure cumulative works, their venues and types, and the full career output trajectory.
Impact—
Insufficient data: citation-based influence signals are not available here beyond h-index (12) and i10-index (12). This axis would measure citations, citing works, and relative influence within the field.
Focus100/100
Focus scores 100, the strongest axis in this profile. The OpenAlex topic distribution is dominated by Topic Modeling with 27 works, followed by Natural Language Processing Techniques (13), Multimodal Machine Learning Applications (12), AI-based Problem Solving and Planning (9), and Artificial Intelligence Applications (7) — five adjacent topics within one NLP/AI cluster. Both reviewed papers are LLM agent evaluation benchmarks, matching the sole fieldDist entry, natural language processing (LLM agent evaluation).
Collaboration60/100
Collaboration scores 60. The recent three-year window records 29 works with 19 distinct coauthors, and twelve recurring collaborators are listed, including Wei Fan, Xinjie Shen, Xudong Guo, Yang Su, and Dayiheng Liu. Both reviewed papers place Tu in a middle author position without corresponding-author status, consistent with a contributing member of larger Alibaba-affiliated teams rather than a solo or lead-coordinator role.
Activity99/100
Activity scores 99, reflecting a steep recent ramp. OpenAlex countsByYear shows 1 work in 2018, 2 in 2020, 1 in 2021, 2 in 2022 and 2 in 2023, before accelerating to 6 works in 2024 and 5 in 2025, with 18 works already recorded in 2026 — the highest single year in the record. The last three years total 29 works, so most recorded output is very recent.
Reproducibility20/100
Reproducibility scores 20 (reproMean 20 across 1 repro-reviewed paper). Evidence is mixed across the two reviewed benchmarks: BabelArena is rated low — no code release, no significance tests, and circular LLM judges — while E-Commerce Bench is rated high for its deterministic, reproducible open design, though 5-run variance and vendor self-evaluation temper its rankings. The low aggregate reflects the small reviewed sample and BabelArena's closed artifacts.
Research summary
Jianhong Tu is a researcher affiliated with Alibaba Group whose OpenAlex-resolved profile centers on natural language processing and, most recently, the evaluation of LLM agents. The bibliometric record shows an h-index of 12 and an i10-index of 12, with a pronounced productivity ramp: recorded yearly output rose from 1 work in 2018 and 2 works in 2020 and 2022 to 6 works in 2024, 5 in 2025, and 18 in 2026, with 29 works in the recent three-year window. Topic Modeling is the largest topic (27 works), followed by Natural Language Processing Techniques (13), Multimodal Machine Learning Applications (12), AI-based Problem Solving and Planning (9), and Artificial Intelligence Applications (7). Evolve has reviewed two 2026 benchmark papers listing Tu as a middle author: BabelArena, a 16,000-instance, 23-language multilingual agent benchmark with low reproducibility readiness, and E-Commerce Bench, a deterministic year-long e-commerce agent benchmark rated high for reproducibility. Nineteen distinct coauthors appear in that window, indicating steady team-based work. Output and impact dimensions lack sufficient data and are left unscored.
Collaboration pattern
Tu's recent work is team-based: 29 works in the last three years involve 19 distinct coauthors, with twelve recurring collaborators listed, including Wei Fan, Xinjie Shen, Xudong Guo, Yang Su, Yinger Zhang, Lianghao Deng, Fengyu Wang, Baohua Dong, Yangqiu Song, Dayiheng Liu, Peng Kuang, and Yuchun Fan. On both reviewed 2026 benchmarks Tu appears in a middle author position without corresponding-author status, indicating a contributing role within larger Alibaba-affiliated teams.
Research keywords
Highlights
BabelArena: a 16k-instance, 23-language benchmark for LLM agents
A 2026 middle-author paper presenting a large-scale multilingual benchmark for tool-using LLM agents, built through a verified translation pipeline covering low-resource languages and probing language consistency. Evolve credits solid engineering but flags no significance tests, circular LLM judges, and no code release, giving it low reproducibility readiness.
E-Commerce Bench: first deterministic year-long benchmark for LLM business agents
A 2026 middle-author paper introducing a deterministic, year-long e-commerce simulation evaluating LLM agents on long-horizon autonomous business operation, including automated negotiation and adversarial fraud. Evolve rates its reproducible open design highly, while noting 5-run variance and vendor self-evaluation temper the rankings.
Timeline
- 2018Earliest year recorded in the OpenAlex countsByYear for this profile, with 1 work.
- 2024Output accelerates: 6 works recorded, up from 2 in 2022 and 2 in 2023.
- 20255 works recorded, sustaining the elevated pace established the previous year.
- 2026Most prolific recorded year with 18 works; two agent benchmarks enter the Evolve catalog — BabelArena and E-Commerce Bench, both with Tu as a middle author.
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 figures come from OpenAlex (identity: openalex) only and may not capture every publication or citation.
- Output and impact dimensions are unscored because the underlying data is insufficient; only h-index (12) and i10-index (12) are available as citation signals.
- The source record lists country as US while the affiliation is Alibaba Group (China); the profile is name-resolved from public bibliographic data and 'Jianhong Tu' may be ambiguous.
- Only the two Evolve-reviewed papers (both 2026) inform the qualitative verdicts; reproducibility aggregates rest on a single repro-reviewed paper.