Research map / Multi-agent systems and games / LLM agents
EpiBench: Benchmarking Multi-turn Research Workflows for Multimodal Agents
Xuan Dong, Huanyang Zheng, Tianhao Niu, Zhe Han, Pengzhan Li, Bofei Liu, Zhengyang Liu, Guancheng Li, Qingfu Zhu, Wanxiang Che
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| enable tool use on the final turn (relax memory-only protocol) (ablation) | LMM research agents · EpiBench | ESR and final-turn accuracy | Improved |
| add PDF text RAG tool (ablation) | LMM research agents · EpiBench | ESR | Mixed |
| raise step budget from 5 to 10 (ablation) | LMM research agents · EpiBench | ESR | Improved |
| raise step budget from 10 to 15 (ablation) | LMM research agents · EpiBench | ESR | No clear effect |
| use thinking mode instead of instruct | Qwen3-VL-235B · EpiBench (hard split) | ESR: 8.16 -> 3.23 | Hurt |
| use thinking mode instead of instruct | Qwen3-VL-235B · EpiBench (average) | final-turn accuracy: 46.08 -> 51.96 | Improved |
| final turn requires memory-based fusion versus earlier turns | GPT-5.2 · EpiBench (hard split) | accuracy: 86.46 -> 49.23 | Hurt |
| GPT-5.2 versus Kimi-K2.5 on evidence correctness | GPT-5.2 · EpiBench (average) | EC: 84.13 -> 86.10 | Improved |
| expert humans versus best agent | Ph.D. experts · EpiBench (hard split) | ESR: 29.23 -> 81.36 | Improved |
Closest papers by meaning
- MiroEval: Benchmarking Multimodal Deep Research Agents in Process and Outcome (NeurIPS 2026 · similarity 0.87)
- AstaBench: Rigorous Benchmarking of AI Agents with a Scientific Research Suite (ICLR 2026 · similarity 0.86)
- ACED-Bench: Evaluating How LLM Agents Acquire and Act on Causal Evidence (NeurIPS 2026 · similarity 0.85)
- ResearchTown: Simulator of Human Research Community (ICML 2025 · similarity 0.85)
- DAGent: Evaluate-then-Grow Planning for Deep Research Agents (NeurIPS 2026 · similarity 0.85)
- SciResearchBench: Benchmarking AI Agents on Complex Scientific Literature Discovery (NeurIPS 2026 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
