Zilong Wang
Publications
CocoaBench: Evaluating Unified Digital Agents in the Wild
LLM agents now perform strongly in software engineering, deep research, GUI automation, and various other applications, while recent agent scaffolds and models are increasingly integrating these capabilities into unified systems. Yet, most evaluations still test these capabilities in isolation, which leaves a gap for more diverse use cases that require agents to combine different capabilities. We introduce CocoaBench, a benchmark for unified digital agents built from human-designed, long-horizon tasks that require flexible composition of vision, search, and coding. Tasks are specified only by an instruction and an automatic evaluation function over the final output, enabling reliable and scalable evaluation across diverse agent infrastructures. We also present CocoaAgent, a lightweight shared scaffold for controlled comparison across model backbones. Experiments show that current agents remain far from reliable on CocoaBench, with the best evaluated system achieving only 45.1% success rate. Our analysis further points to substantial room for improvement in reasoning and planning, tool use and execution, and visual grounding.
Reasoning-Driven Multimodal LLM for Domain Generalization
This paper addresses the domain generalization (DG) problem in deep learning. While most DG methods focus on enforcing visual feature invariance, we leverage the reasoning capability of multimodal large language models (MLLMs) and explore the potential of constructing reasoning chains that derives image categories to achieve more robust predictions under domain shift. To this end, we systematically study the role of reasoning in DG using DomainBed-Reasoning, a newly constructed extension of DomainBed dataset, in which each sample is paired with class-relevant reasoning chains. Our analysis reveals two key challenges: (i) fine-tuning MLLMs with reasoning chains for classification is more challenging than direct label supervision, since the model must optimize complex reasoning sequences before label prediction; and (ii) mismatches in reasoning patterns between supervision signals and fine-tuned MLLMs lead to a trade-off between semantic richness (informative but harder to optimize) and optimization efficiency (easier to optimize but less informative). To address these issues, we propose RD-MLDG (Reasoning-Driven Multimodal LLM for Domain Generalization), a framework with two components: (i) MTCT (Multi-Task Cross-Training), which introduces an additional direct classification pathway to guide reasoning supervision; and (ii) SARR (Self-Aligned Reasoning Regularization), which preserves the semantic richness of reasoning chains while mitigating reasoning-pattern mismatches via iterative self-labeling. Experiments on standard DomainBed datasets (PACS, VLCS, OfficeHome, TerraInc) demonstrate that RD-MLDG achieves state-of-the-art performances, highlighting reasoning as a promising complementary signal for robust out-of-domain generalization.
Mirror: A Multi-Agent System for AI-Assisted Ethics Review
Ethics review is a foundational mechanism of modern research governance, yet contemporary systems face increasing strain as ethical risks arise as structural consequences of large-scale, interdisciplinary scientific practice. The demand for consistent and defensible decisions under heterogeneous risk profiles exposes limitations in institutional review capacity rather than in the legitimacy of ethics oversight. Recent advances in large language models (LLMs) offer new opportunities to support ethics review, but their direct application remains limited by insufficient ethical reasoning capability, weak integration with regulatory structures, and strict privacy constraints on authentic review materials. In this work, we introduce Mirror, an agentic framework for AI-assisted ethical review that integrates ethical reasoning, structured rule interpretation, and multi-agent deliberation within a unified architecture. At its core is EthicsLLM, a foundational model fine-tuned on EthicsQA, a specialized dataset of 41K question-chain-of-thought-answer triples distilled from authoritative ethics and regulatory corpora. EthicsLLM provides detailed normative and regulatory understanding, enabling Mirror to operate in two complementary modes. Mirror-ER (expedited Review) automates expedited review through an executable rule base that supports efficient and transparent compliance checks for minimal-risk studies. Mirror-CR (Committee Review) simulates full-board deliberation through coordinated interactions among expert agents, an ethics secretary agent, and a principal investigator agent, producing structured, committee-level assessments across ten ethical dimensions. Empirical evaluations demonstrate that Mirror significantly improves the quality, consistency, and professionalism of ethics assessments compared with strong generalist LLMs.
Terminal-Bench: Benchmarking Agents on Hard, Realistic Tasks in Command Line Interfaces
AI agents may soon become capable of autonomously completing valuable, long-horizon tasks in diverse domains. Current benchmarks either do not measure real-world tasks, or are not sufficiently difficult to meaningfully measure frontier models. To this end, we present Terminal-Bench 2.0: a carefully curated hard benchmark composed of 89 tasks in computer terminal environments inspired by problems from real workflows. Each task features a unique environment, human-written solution, and comprehensive tests for verification. We show that frontier models and agents score less than 65\% on the benchmark and conduct an error analysis to identify areas for model and agent improvement. We publish the dataset and evaluation harness to assist developers and researchers in future work at https://www.tbench.ai/ .
A Safety Report on GPT-5.2, Gemini 3 Pro, Qwen3-VL, Grok 4.1 Fast, Nano Banana Pro, and Seedream 4.5
The rapid evolution of Large Language Models (LLMs) and Multimodal Large Language Models (MLLMs) has driven major gains in reasoning, perception, and generation across language and vision, yet whether these advances translate into comparable improvements in safety remains unclear, partly due to fragmented evaluations that focus on isolated modalities or threat models. In this report, we present an integrated safety evaluation of six frontier models--GPT-5.2, Gemini 3 Pro, Qwen3-VL, Grok 4.1 Fast, Nano Banana Pro, and Seedream 4.5--assessing each across language, vision-language, and image generation using a unified protocol that combines benchmark, adversarial, multilingual, and compliance evaluations. By aggregating results into safety leaderboards and model profiles, we reveal a highly uneven safety landscape: while GPT-5.2 demonstrates consistently strong and balanced performance, other models exhibit clear trade-offs across benchmark safety, adversarial robustness, multilingual generalization, and regulatory compliance. Despite strong results under standard benchmarks, all models remain highly vulnerable under adversarial testing, with worst-case safety rates dropping below 6%. Text-to-image models show slightly stronger alignment in regulated visual risk categories, yet remain fragile when faced with adversarial or semantically ambiguous prompts. Overall, these findings highlight that safety in frontier models is inherently multidimensional--shaped by modality, language, and evaluation design--underscoring the need for standardized, holistic safety assessments to better reflect real-world risk and guide responsible deployment.
GI-Bench: A Panoramic Benchmark Revealing the Knowledge-Experience Dissociation of Multimodal Large Language Models in Gastrointestinal Endoscopy Against Clinical Standards
Multimodal Large Language Models (MLLMs) show promise in gastroenterology, yet their performance against comprehensive clinical workflows and human benchmarks remains unverified. To systematically evaluate state-of-the-art MLLMs across a panoramic gastrointestinal endoscopy workflow and determine their clinical utility compared with human endoscopists. We constructed GI-Bench, a benchmark encompassing 20 fine-grained lesion categories. Twelve MLLMs were evaluated across a five-stage clinical workflow: anatomical localization, lesion identification, diagnosis, findings description, and management. Model performance was benchmarked against three junior endoscopists and three residency trainees using Macro-F1, mean Intersection-over-Union (mIoU), and multi-dimensional Likert scale. Gemini-3-Pro achieved state-of-the-art performance. In diagnostic reasoning, top-tier models (Macro-F1 0.641) outperformed trainees (0.492) and rivaled junior endoscopists (0.727; p>0.05). However, a critical "spatial grounding bottleneck" persisted; human lesion localization (mIoU >0.506) significantly outperformed the best model (0.345; p<0.05). Furthermore, qualitative analysis revealed a "fluency-accuracy paradox": models generated reports with superior linguistic readability compared with humans (p<0.05) but exhibited significantly lower factual correctness (p<0.05) due to "over-interpretation" and hallucination of visual features. GI-Bench maintains a dynamic leaderboard that tracks the evolving performance of MLLMs in clinical endoscopy. The current rankings and benchmark results are available at https://roterdl.github.io/GIBench/.