An Analysis of Architectural Impact on LLM-based Abstract Visual Reasoning: A Systematic Benchmark on RAVEN-FAIR
This work provides insights into how architectural choices affect LLM performance on visual reasoning tasks, though it is incremental as it benchmarks existing models without introducing new methods.
This study systematically evaluated four large language models on abstract visual reasoning using the RAVEN-FAIR dataset, finding that GPT-4.1-Mini achieved the highest overall accuracy across different reasoning architectures, with results based on best-of-five runs to estimate upper-bound performance.
This study aims to systematically evaluate the performance of large language models (LLMs) in abstract visual reasoning problems. We examined four LLM models (GPT-4.1-Mini, Claude-3.5-Haiku, Gemini-1.5-Flash, Llama-3.3-70b) utilizing four different reasoning architectures (single-shot, embedding-controlled repetition, self-reflection, and multi-agent) on the RAVEN-FAIR dataset. Visual responses generated through a three-stage process (JSON extraction, LLM reasoning, and Tool Function) were evaluated using SSIM and LPIPS metrics; Chain-of-Thought scores and error types (semantic hallucination, numeric misperception) were analyzed. Results demonstrate that GPT-4.1-Mini consistently achieved the highest overall accuracy across all architectures, indicating a strong reasoning capability. While the multi-agent architecture occasionally altered semantic and numeric balance across models, these effects were not uniformly beneficial. Instead, each model exhibited distinct sensitivity patterns to architectural design, underscoring that reasoning effectiveness remains model-specific. Variations in response coverage further emerged as a confounding factor that complicates direct cross-architecture comparison. To estimate the upper-bound performance of each configuration, we report the best of five independent runs, representing a best-case scenario rather than an averaged outcome. This multi-run strategy aligns with recent recommendations, which emphasize that single-run evaluations are fragile and may lead to unreliable conclusions.