ContextMATH reveals LLMs drop 13-34 points on contextual math tasks because they fail at problem formulation, with formulation and reasoning as separate bottlenecks that scale and scenario fine-tuning only partially close.
Here is the feedback: {feedback placeholder} Output your revised version in plain text in the following format: [Step 1] [Mathematical Element]→[Specific Real-World Analogue]
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2026 1verdicts
ACCEPT 1representative citing papers
citing papers explorer
-
From Abstract to Contextual: What LLMs Still Cannot Do in Mathematics
ContextMATH reveals LLMs drop 13-34 points on contextual math tasks because they fail at problem formulation, with formulation and reasoning as separate bottlenecks that scale and scenario fine-tuning only partially close.