Research map / Vision-language and multimodal / Multimodal LLMs
LLaVA-CoT: Let Vision Language Models Reason Step-by-Step
Guowei Xu, Peng Jin, Ziang Wu, Hao Li, Yibing Song, Lichao Sun, Li Yuan
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add structured four-stage reasoning data (vs original Q&A pairs) (ablation) | LLaVA-CoT · Six-benchmark average | average accuracy: 59.0 -> 62.4 | Improved |
| train on original Q&A pairs instead of reasoning data (ablation) | Llama-3.2-11B-Vision-Instruct · MMVet | accuracy: 57.6 -> 49.9 | Hurt |
| add structured stage tags (ablation) | LLaVA-CoT · Six-benchmark average | average accuracy: 60.9 -> 62.4 | Improved |
| add structured stage tags (ablation) | LLaVA-CoT · MMBench V1.1 | accuracy: 74.2 -> 75.0 | Improved |
| add test-time stage-wise retracing search | LLaVA-CoT · Six-benchmark average | average accuracy: 62.4 -> 65.5 | Improved |
| add test-time stage-wise retracing search | LLaVA-CoT · MMStar | accuracy: 57.6 -> 62.5 | Improved |
| train on multi-task sub-skills instead of one CoT chain (ablation) | LLaVA-CoT · Reasoning benchmarks average | average accuracy: 57.7 -> 63.1 | Improved |
| keep natural stage order instead of shuffled order (ablation) | LLaVA-CoT · Reasoning benchmarks average | average accuracy: 58.2 -> 63.1 | Improved |
| prompt base model with structured CoT without training | Llama-3.2-Vision-Instruct · Reasoning benchmarks average | average accuracy: 56.9 -> 56.9 | No clear effect |
| prompt GPT-4o with structured CoT | GPT-4o-0806 · Reasoning benchmarks average | average accuracy: 71.8 -> 74.1 | Improved |
| apply SFT with LLaVA-CoT-100k to base model | Llama-3.2-11B-Vision-Instruct · Six-benchmark average | average accuracy: 56.6 -> 62.4 | Improved |
| apply SFT with LLaVA-CoT-100k to base model | Llama-3.2-11B-Vision-Instruct · MMStar math skill | accuracy: 45.2 -> 64.0 | Improved |
Later papers that compared with this method
7 later accepted papers report LLaVA-CoT as a baseline in their results tables; 7 of them report a better number on most of the results they share with it.
- StaR-KVQA: Structured Reasoning Traces for Implicit-Knowledge Visual Question Answering (CVPR 2026 · better on 2 of 2 shared results)
- Through the Lens of Contrast: Self-Improving Visual Reasoning in VLMs (ICLR 2026 · better on 1 of 1 shared result)
- TumorChain: Interleaved Multimodal Chain-of-Thought Reasoning for Traceable Clinical Tumor Analysis (ICLR 2026 · better on 1 of 1 shared result)
- DRScaffold: Boosting Dense-Scene Reasoning in Lightweight Vision Language Models (NeurIPS 2026 · better on 1 of 1 shared result)
- GThinker: Towards General Multimodal Reasoning via Cue-Guided Rethinking (CVPR 2026 · better on 1 of 1 shared result)
- Integrating Visual Interpretation and Linguistic Reasoning for Geometric Problem Solving (ICCV 2025 · better on 4 of 4 shared results)
- Sherlock: Self-Correcting Reasoning in Vision-Language Models (NeurIPS 2025 · better on 2 of 2 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Visual CoT: Advancing Multi-Modal Language Models with a Comprehensive Dataset and Benchmark for Chain-of-Thought Reasoning (NeurIPS 2024 · similarity 0.91)
- Insight-V: Exploring Long-Chain Visual Reasoning with Multimodal Large Language Models (CVPR 2025 · similarity 0.90)
- HoneyBee: Data Recipes for Vision-Language Reasoners (CVPR 2026 · similarity 0.90)
- Prism: A Framework for Decoupling and Assessing the Capabilities of VLMs (NeurIPS 2024 · similarity 0.90)
- Zebra-CoT: A Dataset for Interleaved Vision-Language Reasoning (ICLR 2026 · similarity 0.88)
- LRM-LLaVA: Overcoming the Modality Gap of Multilingual Large Language-Vision Model for Low-Resource Languages (AAAI 2025 · similarity 0.88)
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.
