Xbench
Emerging8papers using it
2025first seen
XBench is a benchmark used to evaluate expert-level scientific reasoning capabilities of large language models.
Papers using Xbench (8)
- Goal Alignment in LLM-Based User Simulators for Conversational AIScaling Agentic Capabilities via Grounded Interaction SynthesisUncertainty-Aware Clarification in LLM Agents with Information GainWhen Agents Look the Same: Quantifying Distillation-Induced Similarity in Tool-Use BehaviorsLiteResearcher: A Scalable Agentic RL Training Framework for Deep Research AgentReThinker: Scientific Reasoning by Rethinking with Guided Reflection and Confidence ControlScaleEnv: Scaling Environment Synthesis from Scratch for Generalist Interactive Tool-Use Agent TrainingSearch More, Think Less: Rethinking Long-Horizon Agentic Search for Efficiency and Generalization