OpenDataArena/MMFineReason-1.8M-Qwen3-VL-235B-Thinking

MMFineReason是一个大规模、高质量的多模态推理数据集，包含180万样本和51亿解决方案标记，具有从Qwen3-VL-235B-A22B-Thinking提取的详细推理注释。数据集覆盖数学（79.4%）、科学（13.8%）、谜题/游戏（4.6%）和通用/OCR（2.2%）等多个领域。其特点是长形式的链式推理（平均推理长度为2,910标记，是HoneyBee的2.7倍，OpenMMReasoner的4.3倍）和100%的视觉描述覆盖率（平均描述长度为609标记）。数据集通过三阶段构建流程（数据收集与标准化、推理蒸馏和数据选择）确保了高质量和多样性。此外，数据集还提供了详细的统计信息和基准测试结果，展示了其在多模态推理任务中的卓越性能。

MMFineReason is a large-scale, high-quality multimodal reasoning dataset comprising 1.8M samples and 5.1B solution tokens, featuring detailed reasoning annotations distilled from Qwen3-VL-235B-A22B-Thinking. The dataset spans multiple domains: Mathematics (79.4%), Science (13.8%), Puzzle/Game (4.6%), and General/OCR (2.2%). It is characterized by long-form Chain-of-Thought reasoning (average length of 2,910 tokens—2.7× longer than HoneyBee and 4.3× longer than OpenMMReasoner) and 100% caption coverage (average caption length of 609 tokens). The dataset is constructed through a three-stage pipeline (Data Collection & Standardization, Reasoning Distillation, and Data Selection) to ensure high quality and diversity. Additionally, it provides detailed statistics and benchmark results, demonstrating its superior performance in multimodal reasoning tasks.