Sci-MMR: Benchmarking Multi-Step Evidence-Grounded Scientific Reasoning in Multimodal Agents

• URL: http://arxiv.org/abs/2609.11243v1
• Date: Thu, 10 Sep 2026 08:41:07 GMT
• Title: Sci-MMR: Benchmarking Multi-Step Evidence-Grounded Scientific Reasoning in Multimodal Agents
• Abstract summary: Existing multimodal benchmarks largely evaluate final-answer accuracy.<n>We introduce Sci-MMR, a benchmark for multi-step evidence-grounded scientific reasoning.<n>We find that answer accuracy consistently exceeds complete-evidence recovery rate by more than 20%.
• Score: 56.93800079330285
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Autonomous research agents are increasingly expected to search the literature, analyze experimental evidence, and generate scientific hypotheses. These capabilities require multi-step evidence grounded reasoning that progressively acquires, integrates, and verifies evidence before reaching a conclusion. Existing multimodal benchmarks, however, largely evaluate final-answer accuracy, leaving open whether predictions are actually supported by traceable scientific evidence. We introduce Sci-MMR, a benchmark for multi-step evidence-grounded scientific reasoning built on structured argument graphs linking scientific claims, citation-grounded knowledge, visual evidence, and supporting regions. Sci-MMR comprises 235 multi-hop reasoning tasks spanning four scientific disciplines, with an average of nine figure panels per task. Evaluating eight frontier multimodal models, we find that answer accuracy consistently exceeds complete-evidence recovery rate by more than 20%, revealing a substantial gap that answer-only evaluation is structurally unable to capture. Through controlled interventions, we identify two fundamental bottlenecks. First, evidence acquisition: models struggle to extract complete structured evidence from scientific figures, accounting for 57.2% of failures. While cropping tools yield modest gains (+4.5 points), providing gold evidence improves accuracy by up to 37.0 points, indicating difficulty in assembling complete multi-region evidence. Second, evidence integration: models struggle to translate available evidence into correct conclusions, accounting for 31.8% of failures, while even with gold evidence the strongest model achieves only 69.1% accuracy on the hardest tasks. These findings indicate that current answer-centric benchmarks substantially overestimate the evidence-grounded reasoning capabilities of multimodal research agents

Related papers

• Mr.LHDR: A Benchmark for Multimodal Real-World Long-Horizon Deep Research Agents [72.19038675107662]
  Mr.L is a benchmark for evaluating real-world deep research over long, irreducible chains of verifiable evidence.<n>Each question is constructed from a hidden Node-Relation graph and requires an average of 12.1 necessary intermediate conclusions.
  arXiv  Detail & Related papers  (2026-09-10T09:47:44Z)
• NovGauge: A Fine-Grained Benchmark for Diagnosing LLMs' Capability in Paper Novelty Assessment [54.4265627247104]
  Large language models (LLMs) are increasingly used in peer review at major AI conferences.<n>Existing benchmarks assess novelty as a single holistic score.<n>We present NovGauge, a human-anchored benchmark for fine-grained novelty assessment diagnosis.
  arXiv  Detail & Related papers  (2026-09-10T08:34:56Z)
• Not Just Reason, Not Just Scan: Reinforcement Learning for Proactive Scientific Error Verification over Academic Paper [56.13177687090277]
  We present VERA-RL, a reinforcement-learning formulation for scientific error detection over academic papers.<n>We construct VERA-13K, a 12,900-sample dataset organized into 4,300 matched chains, covering 6 scientific-error categories across the research workflow and broad natural-science domains.<n>We further introduce fine-grained rewards for reasoning completeness, evidence alignment, and error precision. Training Qwen3-VL-8B with VERA-RL substantially improves verifiable reasoning, approaching flagship MLLMs such as Gemini 3 Pro and Qwen3-VL-235B-A22B on Scan.
  arXiv  Detail & Related papers  (2026-08-27T04:24:29Z)
• TRACES: A Benchmark for Epistemic Reliability in Scientific Reasoning by LLMs [0.0]
  Large language models are being proposed as agents in scientific, in domains where no downstream verifier exists.<n>We introduce a probe corpus of 42 retracted, fraudulent, and pseudoscientific papers, paired with a methodology for eliciting and scoring single-shot model engagement with each paper's framing.<n>The probes span five claim types: fabricated observation, pseudophysical mechanism, magical premise, legitimization bridge, and cargo-cult experiment.
  arXiv  Detail & Related papers  (2026-08-11T20:30:32Z)
• HiEviDR-Bench: A Benchmark for Hierarchical Evidence Aggregation in Deep Research [52.87128503345243]
  We introduce HiEviDR-Bench, a benchmark for evaluating Hierarchical Evidence Aggregation in Deep Research.<n>HiEviDR-Bench represents each instance with an explicit evidence graph that captures evidence selection, cross-source linking, and aggregation.<n>We develop a traceability-oriented evaluation framework with five dimensions: report quality, evidence traceability, citation accuracy, claim verification, and answer correctness.
  arXiv  Detail & Related papers  (2026-07-27T23:50:46Z)
• AI scientists produce results without reasoning scientifically [3.100302590436282]
  Large language model (LLM)-based systems are increasingly deployed to conduct scientific research autonomously.<n>Here, we evaluate LLM-based scientific agents across eight domains, spanning execution to hypothesis-driven inquiry.
  arXiv  Detail & Related papers  (2026-04-20T20:23:42Z)
• M2-Verify: A Large-Scale Multidomain Benchmark for Checking Multimodal Claim Consistency [17.049806985851994]
  M2-Verify is a large-scale multimodal dataset for checking scientific claim consistency.<n>It provides over 469K instances across 16 domains, rigorously validated through expert audits.
  arXiv  Detail & Related papers  (2026-04-01T18:18:10Z)
• MultiHaystack: Benchmarking Multimodal Retrieval and Reasoning over 40K Images, Videos, and Documents [57.32877731797049]
  MultiHaystack is the first benchmark designed to evaluate both retrieval and reasoning under large-scale, cross-modal conditions.<n>We find that models perform competitively when provided with the corresponding evidence, but their performance drops sharply when required to retrieve that evidence from the full corpus.
  arXiv  Detail & Related papers  (2026-03-05T21:43:02Z)
• Can AI Validate Science? Benchmarking LLMs for Accurate Scientific Claim $\rightarrow$ Evidence Reasoning [6.043212666944194]
  We present CLAIM-BENCH, a benchmark for evaluating large language models' capabilities in scientific claim-evidence extraction and validation.<n>We show that closed-source models like GPT-4 and Claude consistently outperform open-source counterparts in precision and recall.<n> strategically designed three-pass and one-by-one prompting approaches significantly improve LLMs' abilities to accurately link dispersed evidence with claims.
  arXiv  Detail & Related papers  (2025-06-09T21:04:39Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.