mlsanityData to 14 Oct 2026

π2: Structure-Originated Reasoning Data Improves Long-Context Reasoning Ability of Large Language Models

arXiv 2604.05114 ↗First posted 6 Apr 2026cs.CL cs.AI cs.LGcited by 0

On the map: LLM reasoning → Reasoning benchmarks and evaluation

Abstract (arXiv)

We study a QA curation pipeline for improving long-context complex reasoning in large language models (LLMs). Our approach, π2, constructs high-quality reasoning data through rigorous QA curation: 1) extracting and expanding tables from Wikipedia, 2) from the collected tables together with relevant metadata, generating complex reasoning questions whose answers are automatically determined and validated through dual-path code execution, 3) finally, back-translating chain-of-thoughts solutions grounded in realistic context. Supervised fine-tuning with gpt-oss-20b and Qwen3-4B-Instruct-2507 on π2 yields consistent improvements across four long-context reasoning benchmarks and our alike π2-Bench, with average absolute accuracy gains of +6.25% and +3.37% respectively. Through deeper analyses, we observe that reasoning style contributes little, while faithful reasoning patterns discovered by back translation and grounded realistic long context, as π2 is designed for, are crucial for the improvement. Our code, data, and models are fully open-source at https://github.com/vtpss/pi-squared.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning. Builds on 29 of its 85 references that are in the map. First posted on arXiv 6 Apr 2026.

Builds on

29 of 85 references.

Closest ancestors · 5

Same direction, newest first

  1. 2025OpenThoughts: Data Recipes for Reasoning Models4 Jun
  2. 2025Qwen3 Technical Report14 May
  3. 2025LongFaith: Enhancing Long-Context Reasoning in LLMs with Faithful Synthetic Data18 Feb
  4. 2024LongBench v2: Towards Deeper Understanding and Reasoning on Realistic Long-context Multitasks19 Dec
  5. 2024BABILong: Testing the Limits of LLMs with Long Context Reasoning-in-a-Haystack14 Jun

Other references · 24

  1. 2021LoRA: Low-Rank Adaptation of Large Language Modelscited by 14,380
  2. 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,944
  3. 2018HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answeringcited by 3,754
  4. 2023LongBench: A Bilingual, Multitask Benchmark for Long Context Understandingcited by 1,417
  5. 2023MemGPT: Towards LLMs as Operating Systemscited by 1,389
  6. 2025s1: Simple test-time scalingcited by 1,341
  7. 2021MuSiQue: Multihop Questions via Single-hop Question Compositioncited by 916
  8. 2023MemoryBank: Enhancing Large Language Models with Long-Term Memorycited by 556
  9. 2025LIMO: Less is More for Reasoningcited by 498
  10. 2023L-Eval: Instituting Standardized Evaluation for Long Context Language Modelscited by 233
  11. 2025BrowseComp-Plus: A More Fair and Transparent Evaluation Benchmark of Deep-Research Agentcited by 200
  12. 2024Make Your LLM Fully Utilize the Contextcited by 119
12 more
  1. 2025Small Models Struggle to Learn from Strong Reasonerscited by 113
  2. 2025Recursive Language Modelscited by 104
  3. 2025SealQA: Raising the Bar for Reasoning in Search-Augmented Language Modelscited by 75
  4. 2024DocFinQA: A Long-Context Financial Reasoning Datasetcited by 60
  5. 2025Oolong: Evaluating Long Context Reasoning and Aggregation Capabilitiescited by 40
  6. 2026OfficeQA Pro: An Enterprise Benchmark for End-to-End Grounded Reasoningcited by 39
  7. 2024FanOutQA: A Multi-Hop, Multi-Document Question Answering Benchmark for Large Language Modelscited by 29
  8. 2025LoongRL: Reinforcement Learning for Advanced Reasoning over Long Contextscited by 29
  9. 2025QwenLong-L1.5: Post-Training Recipe for Long-Context Reasoning and Memory Managementcited by 27
  10. 2024What are the Essential Factors in Crafting Effective Long Context Multi-Hop Instruction Datasets? Insights and Best Practicescited by 14
  11. 2025CLIPPER: Compression enables long-context synthetic data generationcited by 10
  12. 2026Probing How Scalable Table Data Enhances General Long-Context Reasoningcited by 1

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025LongReason: A Synthetic Long-Context Reasoning Benchmark via Context Expansion
  2. 2024Michelangelo: Long Context Evaluations Beyond Haystacks via Latent Structure Queries
  3. 2024Large Language Models Can Self-Improve in Long-context Reasoning
  4. 2026A Decomposition Perspective to Long-context Reasoning for LLMs
  5. 2024ALR2: A Retrieve-then-Reason Framework for Long-context Question Answering
  6. 2025InftyThink: Breaking the Length Limits of Long-Context Reasoning in Large Language Models
  7. 2025Follow the Path: Reasoning over Knowledge Graph Paths to Improve Large Language Model Factuality
  8. 2025MobileLLM-R1: Exploring the Limits of Sub-Billion Language Model Reasoners with Open Training Recipes