AI Paper Atlas

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

From Abstract to Contextual: What LLMs Still Cannot Do in Mathematics

Bowen Cao, Dongdong Zhang, Yixia Li, Junpeng Liu, Shijue Huang, Chufan Shi, Hongyuan Lu, Yaokang Wu, Guanhua Chen, Wai Lam, Furu Wei

ICLR 2026 · Large language models › Agents, code and math · approach: Mathematical and structural reasoning benchmarks · 2 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
embed AIME problems in realistic scenarios (SG) and sub-problem conditions (CS)Qwen3-32B · AIME 2024accuracy: 81.25 -> 67.92 (SG), 57.08 (CS)Hurt
embed AIME 2025 problems in SG and CS scenariosDeepSeek-R1 · AIME 2025accuracy: 86.67 -> 73.33 (SG), 53.33 (CS)Hurt
increase model size within OpenMath-Nemotron seriesOpenMath-Nemotron · AIME 2024 CSrelative drop: 77% (1.5B) -> 43% (14B), 51% (32B)Improved
further RL specialization from R1-Distil-Qwen-7B to AceMath-RL-Nemotron-7BR1-Distil-Qwen-7B · AIME 2024 SGrelative drop: -18% -> -30%Hurt
further training from Qwen3-8B to R1-0528-Qwen3-8BQwen3-8B · AIME 2024 CSaccuracy: 42.9 -> 39.6Hurt
increase problem difficulty from MATH to AIME24 to AIME25Qwen3-4B · formulationformulation accuracy: 84.8 -> 64.6 -> 47.1Hurt
scale model from Qwen3-0.6B to Qwen3-32BQwen3 · formulationavg formulation accuracy: 42.8 -> 75.0Improved
add scenario-only SFT instead of original-only SFT (ablation)Qwen3-8B-Base · ContextMATH AIME 2024 SGaccuracy: 35.4 -> 44.8Improved
add scenario-only SFT instead of original-only SFT (ablation)Qwen3-14B-Base · ContextMATH AIME 2024 CSaccuracy: 25.2 -> 38.8Improved
add scenario SFT to base model (mixture regime) (ablation)Qwen3-14B-Base · AIME 2024 SGaccuracy: 11.0 -> 52.5Improved
mix original and scenario data instead of original only (ablation)Qwen3-8B-Base · average over benchmarksaverage accuracy: 52.1 -> 55.6Improved
scenario SFT on abstract AMC23 (ablation)Qwen3-8B-Base · AMC23accuracy: 83.9 (Ori) -> 83.3 (Syn)No 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