AI Paper Atlas

Research map / Large language models / Agents, code and math

MME-Reasoning: A Broad-Spectrum Benchmark for Evaluating Logical Reasoning in MLLMs

Jiakang Yuan, Tianshuo Peng, Yilei Jiang, Yiting Lu, Renrui Zhang, Kaituo Feng, Chaoyou Fu, Tao Chen, LEI BAI, Bo Zhang

ICML 2026 · Large language models › Agents, code and math · approach: Reasoning benchmarks and platforms

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use thinking-mode variant of the model (ablation)Seed1.5-VL · MME-Reasoningaccuracy (AVG): 47.5 -> 59.9Improved
train 7B models with rule-based RL (thinking activation)7B base MLLMs · MME-Reasoningaccuracy (AVG)Hurt
add MCTS test-time scaling with a 32B reward model (ablation)Qwen2.5-VL-7B · MME-Reasoningaccuracy (AVG): 26.8 -> 23.3Hurt
add CoT prompt (ablation)Qwen2.5-VL-7B · MME-Reasoningaccuracy (AVG): 26.8 -> 24.7Hurt
add CoT prompt (ablation)Qwen2.5-VL-32B · MME-Reasoningaccuracy (AVG): 33.2 -> 32.3No clear effect
add CoT prompt (ablation)Qwen2.5-VL-72B · MME-Reasoningaccuracy (AVG): 34.1 -> 33.0No clear effect
add CoT prompt (ablation)InternVL3-38B · MME-Reasoningaccuracy (AVG): 28.4 -> 32.7Improved
add CoT prompt (ablation)InternVL3-78B · MME-Reasoningaccuracy (AVG): 32.1 -> 32.9No clear effect
add CoT prompt (ablation)InternVL3-8B · MME-Reasoningaccuracy (AVG): 26.4 -> 25.2Hurt
longer output (more tokens) across modelsMLLMs · MME-ReasoningaccuracyMixed
replace image with GPT-4o caption and use thinking LLMDeepSeek-R1 · MME-Reasoningaccuracy (AVG): 46.9 (R1) vs 34.1 (Claude-3.7-Sonnet-Thinking)Improved
increase question difficulty levelmodels · MME-ReasoningaccuracyHurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy