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

arXiv: https://arxiv.org/abs/2609.11243 · HTML: https://mlsanity.com/paper/2609.11243/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Knowledge- and tool-augmented reasoning, one of 17 directions in LLM reasoning.
- Builds on 15 of its 37 references that are in the map.
- First posted on arXiv 10 Sep 2026.

## Placed in

- LLM reasoning / [Knowledge- and tool-augmented reasoning](https://mlsanity.com/lr/knowledge-and-tool-augmented-reasoning/)
- LLM agents and tool use / [Web, search, and research agents](https://mlsanity.com/agents/web-search-and-research-agents/)

## Builds on (most cited first, 15 of 15)

- [The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery](https://mlsanity.com/paper/2408.06292/) (2024)
- [Accelerating scientific discovery with Co-Scientist](https://mlsanity.com/paper/2502.18864/) (2025)
- [MinerU: An Open-Source Solution for Precise Document Content Extraction](https://mlsanity.com/paper/2409.18839/) (2024)
- [WebSailor: Navigating Super-human Reasoning for Web Agent](https://mlsanity.com/paper/2507.02592/) (2025)
- [ResearchRubrics: A Benchmark of Prompts and Rubrics For Evaluating Deep Research Agents](https://mlsanity.com/paper/2511.07685/) (2025)
- [Vision-DeepResearch Benchmark: Rethinking Visual and Textual Search for Multimodal Large Language Models](https://mlsanity.com/paper/2602.02185/) (2026)
- [GLM-5V-Turbo: Toward a Native Foundation Model for Multimodal Agents](https://mlsanity.com/paper/2604.26752/) (2026)
- [MMDeepResearch-Bench: A Benchmark for Multimodal Deep Research Agents](https://mlsanity.com/paper/2601.12346/) (2026)
- [MC-Search: Evaluating and Enhancing Multimodal Agentic Search with Structured Long Reasoning Chains](https://mlsanity.com/paper/2603.00873/) (2026)
- [Towards Long-horizon Agentic Multimodal Search](https://mlsanity.com/paper/2604.12890/) (2026)
- [SciVisAgentBench: A Benchmark for Evaluating Scientific Data Analysis and Visualization Agents](https://mlsanity.com/paper/2603.29139/) (2026)
- [MiroEval: Benchmarking Multimodal Deep Research Agents in Process and Outcome](https://mlsanity.com/paper/2603.28407/) (2026)
- [HopChain: Multi-Hop Data Synthesis for Generalizable Vision-Language Reasoning](https://mlsanity.com/paper/2603.17024/) (2026)
- [CRIT: Graph-Based Automatic Data Synthesis to Enhance Cross-Modal Multi-Hop Reasoning](https://mlsanity.com/paper/2604.01634/) (2026)
- [M2-Verify: A Large-Scale Multidomain Benchmark for Checking Multimodal Claim Consistency](https://mlsanity.com/paper/2604.01306/) (2026)

## Abstract (arXiv)

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