AI agents promise to turn scattered scientific evidence into AI-ready datasets, yet existing benchmarks examine isolated NLP skills or paper-level summarization and leave open whether agents can assemble complete, evidence-grounded records across multimodal and heterogeneous scientific sources. To address this gap, we introduce ABDataBench, a general evaluation protocol for antibody functional data construction from scientific literature and patents. ABDataBench contains 44 expert-annotated, OCR-processed documents, including 27 papers and 17 patents, yielding 279 antibody-level records under a unified 22-field schema. We further develop ABCurator, a record-centric multi-agent framework that treats each antibody as the unit for planning, evidence aggregation, enrichment, and validation, coupled with a skill-accumulation workflow that distills expert feedback into reusable agent competencies across evaluation rounds. Across proprietary and open-source models, the strongest configurations achieve scores of 78.3 and 76.0, respectively. Analysis shows that current agents achieve high antibody recall but remain limited by over-generation, sequence attribution errors, and incomplete affinity grounding. These results position antibody record construction as a challenging long-horizon information-extraction task and establish ABDataBench as a benchmark framework for reliable AI-driven scientific data generation. The source code and data for ABDataBench is released at:https://github.com/GAIR-NLP/ABDataBench.

Large Language Models (LLMs) based autonomous agents demonstrate multifaceted capabilities to contribute substantially to economic production. However, existing benchmarks remain focused on single agentic capability, failing to capture long-horizon real-world scenarios. Moreover, the reliance on human-in-the-loop feedback for realistic tasks creates a scalability bottleneck, hindering automated rollout collection and evaluation. To bridge this gap, we introduce AgencyBench, a comprehensive benchmark derived from daily AI usage, evaluating 6 core agentic capabilities across 32 real-world scenarios, comprising 138 tasks with specific queries, deliverables, and rubrics. These scenarios require an average of 90 tool calls, 1 million tokens, and hours of execution time to resolve. To enable automated evaluation, we employ a user simulation agent to provide iterative feedback, and a Docker sandbox to conduct visual and functional rubric-based assessment. Experiments reveal that closed-source models significantly outperform open-source models (48.4% vs 32.1%). Further analysis reveals significant disparities across models in resource efficiency, feedback-driven self-correction, and specific tool-use preferences.

High-quality scientific data is critical for advancing LLMs, yet academic literature remains largely underutilized. This work addresses the fundamental question: How can we systematically unlock scientific data’s value for pre-training? First, we construct a large-scale raw scientific corpus but identify a critical Learnability Gap, revealing that direct pre-training yields negligible gains. To bridge this, we develop a multi-stage pipeline featuring content cleaning and pedagogical augmentation, resulting in SciPedia, a 900B-token corpus. Finally, we establish a controlled verification framework: we develop SciPedia-Eval benchmark and conduct 600B tokens of continued pre-training (CPT) starting from transparent base models (3B/7B) trained from scratch. Compared to a CPT baseline trained with general-purpose data, our approach with SciPedia data boosts average performance by +2.12 (3B) and +2.95 (7B), reaching +5.60 and +8.40 on in-domain tasks. This setup further allows us to derive empirical guidelines for data composition and model configurations.