Google Scholar
GitHub
X
Hugging Face
Email
RL for Reasoning
Post-training recipes for math, STEM, code, and long-horizon reasoning models.
Chenchen Zhang
Senior Researcher, Tencent · Hunyuan Department
I am a Senior Researcher at Tencent Hunyuan, where I work on RL post-training and agentic code-reasoning systems for foundation models. I have been a core contributor to Hy3.0 preview, HY2.0, Hunyuan-A13B, Hunyuan-TurboS, Hunyuan-T1, and Hunyuan-Large.
Current Focus
What I am working on now
Agentic RL
Training loops for web coding agents, tool use, critic feedback, and credit assignment.
Code Intelligence
Repository-level generation, automatic benchmark generation, visual-to-code, and code critique.
Multimodal Evaluation
Evaluation for audio-visual understanding, artifacts, tool-use, and model iteration loops.
Hunyuan
Model Family Timeline
A compact view of Hunyuan model releases and technical reports where I have been a core contributor.
Hunyuan-Large
Open-source MoE model with long-context support.
Hunyuan-T1
Reasoning-focused model with code-reasoning optimization and code evaluation.
Hunyuan-TurboS & A13B
Efficient reasoning releases: adaptive CoT in TurboS and open-source MoE A13B with hybrid reasoning.
HY2.0
Think and Instruct variants for reasoning and instruction-following.
Hy3.0 preview
Open-source preview model with reasoning, coding, and agentic workflows.
Research
Research Directions
I am currently working on foundation models, especially code/reasoning models, reinforcement learning, agentic RL, math and STEM reasoning, code intelligence, multimodal systems, and evaluation.
Foundation
Hunyuan Models
Hy3.0 · HY2.0 · A13B · T1 · TurboS · Large
Optimization
Math / STEM RL
CriticLean · ConceptMath · RLPT
Agents
Agentic RL
ReLook · Rhombus · Credit Assignment
Code
Code Intelligence
SWE-Compass · NL2Repo · OpenCoder
Evaluation
Multimodal & Tool Use
OmniVideoBench · MTU-Bench · ArtifactsBench
Pretraining & Foundation Models
Hy3.0 preview, HY2.0, Hunyuan-A13B, Hunyuan-TurboS, Hunyuan-T1, Hunyuan-Large, MAP-Neo, OpenCoder, D-CPT Law, E2-LLM, DDK.
Reasoning & Agentic RL
Math/STEM reasoning optimization, RL for code and reasoning models, reinforcement learning on pre-training data, critic-guided RL, parallel thinking, and credit assignment.
Code & Agentic LLMs
Agentic coding evaluation, long-horizon repository generation, automatic benchmark generation, visual-to-code, code critique, artifact evaluation, and code LLM pretraining.
Evaluation, Multimodal & Surveys
Math, code, tool-use, audio-visual, multimodal browsing, group identity evaluation, and survey resources.
Updates
News
Selected paper releases, conference updates, and open-source project milestones.
-
ICML 2026: SWE-Compass, NL2Repo-Bench, and From Diagrams to Code.
-
ACL 2026: CriticLean, RLPT, ReLook, and Rhombus.
-
Hy3.0 preview open-sourced; #1 on OpenRouter daily global API usage.
-
ICLR 2026: AutoCodeBench and OmniVideoBench.
-
Core contributor to the HY2.0 model family.
-
Core contributor to open-source Hunyuan-A13B, an efficient MoE model with hybrid reasoning and agent capabilities.
-
Hunyuan-TurboS technical report released.
-
ACL 2025: OpenCoder.
-
Core contributor to reasoning-focused Hunyuan-T1.
-
ICLR 2025: MTU-Bench.
-
Hunyuan-Large weights and technical report released.
-
ACL 2024: E2-LLM and ConceptMath.
Selected Work
Selected Publications & Projects
Full list on Google Scholar Full listRepresentative models, papers, benchmarks, and survey resources.
Models & Technical Reports
Hy3.0 preview
Open Model 2026Open-source Hunyuan 3.0 preview model for reasoning, coding, and agentic workflows.
My role: Core contributor to post-training, RL recipes, and reasoning optimization across STEM and code.
HY2.0
Model Family 2025Next-generation Hunyuan model family for reasoning and instruction-following scenarios.
My role: Core contributor to post-training and evaluation for reasoning and instruction-following models.
Hunyuan-A13B: An Efficient Open-Source MoE Model with 13B Active Parameters
Open Model 2025Open-source fine-grained MoE model with 80B total parameters, 13B active parameters, hybrid reasoning, long-context support, and strong agent capabilities.
My role: Core author and contributor, with contribution level comparable to Hunyuan-T1.
Hunyuan-TurboS: Advancing Large Language Models through Mamba-Transformer Synergy and Adaptive Chain-of-Thought
Tech Report 2025My role: Core contributor to reasoning model optimization and adaptive CoT evaluation.
Hunyuan-T1
Tech Page 2025My role: Core contributor to reasoning post-training, code-reasoning optimization, LiveCodeBench SOTA, and code evaluation.
Hunyuan-Large: An Open-Source MoE Model with 52 Billion Activated Parameters by Tencent
arXiv 2024Open-source MoE LLM with long-context support.
My role: Core contributor to model development and evaluation for the Hunyuan open-source LLM family.
Code & Agentic LLMs
SWE-Compass: Towards Unified Evaluation of Agentic Coding Abilities for Large Language Models
ICML 2026* Equal contribution. My role: Built unified evaluation for agentic coding abilities and code-agent capability analysis.
NL2Repo-Bench: Towards Long-Horizon Repository Generation Evaluation of Coding Agents
ICML 2026From Diagrams to Code: Multilingual Programming with Visual Design
ICML 2026ReLook: Vision-Grounded RL with a Multimodal LLM Critic for Agentic Web Coding
ACL 2026* Equal contribution. My role: Designed agentic RL evaluation and multimodal critic loops for web coding.
WebCompass: Towards Multimodal Web Coding Evaluation for Code Language Models
arXiv 2026* Equal contribution. My role: Built multimodal web-coding evaluation and analysis for stronger code and agentic capabilities.
AutoCodeBench: Large Language Models are Automatic Code Benchmark Generators
ICLR 2026OpenCoder: The Open Cookbook for Top-Tier Code Large Language Models
ACL 2025ArtifactsBench: Bridging the Visual-Interactive Gap in LLM Code Generation Evaluation
arXiv 2025Automated multimodal evaluation for interactive visual artifacts.
My role: First author; built visual-interactive code evaluation for artifact generation.
CodeCriticBench: A Holistic Code Critique Benchmark for Large Language Models
arXiv 2025My role: First author; built holistic critique evaluation for code LLMs and critic-model analysis.
RL & Reasoning
CriticLean: Critic-Guided Reinforcement Learning for Mathematical Formalization
ACL 2026Reinforcement Learning on Pre-Training Data
ACL 2026Rhombus: Incentivizing Coordination in Parallel Thinking through Reinforcement Learning
ACL 2026From Reasoning to Agentic: Credit Assignment in Reinforcement Learning for Large Language Models
arXiv 2026Survey and resource hub for reinforcement learning credit assignment in LLMs and agents.
Foundation, Pretraining & Efficiency
MAP-Neo: Highly Capable and Transparent Bilingual Large Language Model Series
arXiv 2024DDK: Distilling Domain Knowledge for Efficient Large Language Models
NeurIPS 2024* Equal contribution. My role: Co-first author.
D-CPT Law: Domain-specific Continual Pre-Training Scaling Law for Large Language Models
NeurIPS 2024My role: Core author; wrote most of the underlying codebase for domain-specific continual pre-training and scaling-law experiments.
E2-LLM: Efficient and Extreme Length Extension of Large Language Models
ACL 2024Multimodal, Evaluation & Surveys
OmniVideoBench: Towards Audio-Visual Understanding Evaluation for Omni MLLMs
ICLR 2026MTU-Bench: A Multi-granularity Tool-Use Benchmark for Large Language Models
ICLR 2025ConceptMath: A Bilingual Concept-wise Benchmark for Mathematical Reasoning of LLMs
ACL 2024GIEBench: Towards Holistic Evaluation of Group Identity-based Empathy for Large Language Models
EMNLP 2024A Comprehensive Survey on Long Context Language Modeling
SurveyBackground
Background
Research, AI infrastructure, and engineering roles across large-scale language model systems.
Experience
Tencent
Tencent
Hunyuan Department · Senior Researcher
Alibaba
Alibaba
Algorithm Engineer · AI Infrastructure & Research, with early work focused on AI infra
Alibaba
Algorithm Engineer Intern · AI Infrastructure & research systems
Baidu
Baidu
Algorithm Engineer Intern