LLM reasoning / chain-of-thought
LLaVA-CoT
LLaVA-CoT: Let Vision Language Models Reason Step-by-Step
Superseded baseline#55 of 772 most-superseded · first seen Nov 15, 2024
Superseded — cited as a baseline and beaten by newer methods
1 papers critique it · 1 beat it on benchmarks
What papers say
Verbatim critique sentences, each from a paper that cites LLaVA-CoT as a baseline.
However, these approaches, even advanced models like GPT-5 or Gemini, perform CoT in pure text space. Once visual features are initially encoded, they cannot be re-accessed during reasoning.
Beaten on benchmarks
Head-to-head results where a newer method reports beating LLaVA-CoT. Values are copied from the source paper's tables — verify against the cited paper.
TVI-CoT beats LLaVA-CoT
73.2 vs 48.6
Figure question answering accuracy · [MathVista]
TVI-CoT: Text-Visual Interleaved Chain-of-Thought Reasoning for Multimodal Understanding
What to use instead
Recent methods in the same sub-problem, not yet superseded in the knowledge base — arXiv benchmark leaders, not vetted production recommendations.
- Jun 7, 2026