AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

AtomThink: Multimodal Slow Thinking With Atomic Step Reasoning

Kun Xiang, Zhili Liu, Terry Jingchen Zhang, Yinya Huang, Yunshuang Nie, Kaixin Cai, Yiyang Yin, Runhui Huang, Hanhui Li, Yihan Zeng, Yu-Jie Yuan, Jianhua Han and 3 more

TPAMI 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Chain-of-thought multimodal reasoning · 28 citations (OpenAlex)

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
fine-tune on AMATH-SFT and use SCoT inference (ablation)Llama3.2-Vision-11B · MathVistaaccuracy (%): 44.3 -> 57.1Improved
use Self-structured CoT inference instead of standard CoTAtomThink-LlamaV · MathVistaaccuracy (%): 50.4 -> 57.1Improved
add PRM-guided step-wise beam search to SCoTAtomThink-LlamaV · MathVistaaccuracy (%): 57.1 -> 58.4Improved
add PRM-guided search to SCoTAtomThink-LLaVA · WeMathaccuracy (%): 34.5 -> 44.2Improved
add PRM-guided search to SCoTAtomThink-LlamaV · ScienceQAaccuracy (%): 85.9 -> 85.8Hurt
use beam search instead of greedy step-wise search (ablation)AtomThink-EMOVA · MathVerse (300-sample subset)accuracy (%): 38.3 -> 45.3Improved
use MCTS instead of beam search (ablation)AtomThink-EMOVA · MathVerse (300-sample subset)accuracy (%): 45.3 -> 47.6Improved
use Best-of-N Avg aggregation instead of SCoT alone (ablation)AtomThink-EMOVA · MathVista-M vs MathVista-Gaccuracy (%): 54.2 -> 58.7 on MathVista-M; 46.7 -> 40.5 on MathVista-GMixed
increase beam-search candidates from 0 to 3 (ablation)AtomThink-LlamaV · MathVision-miniaccuracy (%): 13.9 -> 23.3Improved
scale AMATH-SFT data from 0k to 124k (ablation)AtomThink-LLaVA · MathVision-miniaccuracy (%): 9.28 -> 12.45Improved
add 100k LLaVA-Instruct data instead of none (ablation)AtomThink-LLaVA · MathVision-miniaccuracy (%): 8.81 -> 12.45Improved
scale LLaVA-Instruct from 100k to 400k (ablation)AtomThink-LLaVA · MathVision-miniaccuracy (%): 12.45 -> 12.50No clear effect

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