Compliance, Capability, and Conflict: Benchmarking Multimodal LLMs under System Messages
On the map: Multimodal LLMs / vision-language models → MLLM training and visual instruction tuning
Abstract (arXiv)
Production deployments of Multimodal Large Language Models (MLLMs) increasingly rely on system messages to govern model behavior. Yet existing benchmarks either evaluate constraints in text only or embed them into the user turn, leaving system-message adherence in multimodal contexts largely unmeasured; they also leave open whether compliance comes at the cost of foundational vision-language capabilities. We introduce VSysBench, a benchmark built on MMVet-v2 that organizes constraints into 5 main categories and 22 sub-categories, ranging from textual directives in visual contexts to fully vision-grounded ones, each paired with a misaligned counterpart that stress-tests the instructional hierarchy. VSysBench scores each response jointly along two axes, constraint compliance and answer correctness, via the Joint Satisfaction Rate (JSR) and Cross-Constraint Sensitivity (CCS). Across 16 MLLMs, we find that imposing system messages substantially erodes base task accuracy, that compliance collapses under user conflict for open-weight models while remaining stable for top proprietary ones, and that vision-grounded constraints are the hardest category for every model.
Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in MLLM training and visual instruction tuning, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 17 of its 26 references that are in the map. First posted on arXiv 12 Jun 2026.
Builds on
17 of 26 references.
Closest ancestors · 8
Same direction, newest first
- 2026Empowering Reliable Visual-Centric Instruction Following in MLLMs6 Jan
- 2025MM-IFEngine: Towards Multimodal Instruction Following10 Apr
- 2024LLaVA-OneVision: Easy Visual Task Transfer6 Aug
- 2024MM-Vet v2: A Challenging Benchmark to Evaluate Large Multimodal Models for Integrated Capabilities1 Aug
- 2024MIA-Bench: Towards Better Instruction Following Evaluation of Multimodal LLMs1 Jul
- 2023Improved Baselines with Visual Instruction Tuning5 Oct
- 2023MM-Vet: Evaluating Large Multimodal Models for Integrated Capabilities4 Aug
- 2023MMBench: Is Your Multi-modal Model an All-around Player?12 Jul
Other references · 9
- 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,626
- 2024The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructionscited by 493
- 2024Benchmarking Complex Instruction-Following with Multiple Constraints Compositioncited by 152
- 2024Aligning to Thousands of Preferences via System Message Generalizationcited by 84
- 2024FOFO: A Benchmark to Evaluate LLMs' Format-Following Capabilitycited by 80
- 2023Can Large Language Models Understand Real-World Complex Instructions?cited by 69
- 2024Measuring and Controlling Instruction (In)Stability in Language Model Dialogscited by 53
- 2024SysBench: Can Large Language Models Follow System Messages?cited by 28
- 2024SoFA: Shielded On-the-fly Alignment via Priority Rule Followingcited by 18
Built on by
Not cited yet.
Similar, not cited
Close in wording, no citation either way
- 2026Same Semantics, Different Outcome: On the Modality Robustness of Multimodal LLMs under Knowledge Conflict
- 2026CSR-Bench: A Benchmark for Evaluating the Cross-modal Safety and Reliability of MLLMs
- 2025Can Large Multimodal Models Actively Recognize Faulty Inputs? A Systematic Evaluation Framework of Their Input Scrutiny Ability
- 2025Hidden in Plain Sight: Reasoning in Underspecified and Misspecified Scenarios for Multimodal LLMs
- 2026MMPCBench: Benchmarking Multimodal Large Language Models on Proactive Critique of Flawed Inputs
- 2024Insight Over Sight: Exploring the Vision-Knowledge Conflicts in Multimodal LLMs
- 2023Beyond Task Performance: Evaluating and Reducing the Flaws of Large Multimodal Models with In-Context Learning
- 2025Can't See the Forest for the Trees: Benchmarking Multimodal Safety Awareness for Multimodal LLMs