Data World: Scaling Data Environments from Scratch

Hugging Face Daily Papers Papers
multi-agent-systems data-analysis reinforcement-learning benchmark synthetic-data agentic-ai

Summary

论文提出 Data World，一个多智能体引擎，从显式机制出发合成数据环境、任务与评分标准，用于训练数据智能体的分析洞察能力；基于其合成任务通过 RL-zero 训练得到 DataWorld-35B-A3B 模型，在 DataWorld-Bench、DAComp、DABstep-Hard 与 JobBench 等基准上显著超越基座模型。

Data agents use code to process heterogeneous files and derive analytical insights. Scaling their training faces costly environment curation, privacy and intellectual-property restrictions, and difficulty verifying open-ended findings without known generative truth. Our core insight is to specify the world behind the files before generating the files themselves. We introduce Data World, a multi-agent engine that synthesizes environments, tasks, and rubrics from explicit mechanisms. Independent audits check target recoverability from observed files; rubrics support evaluation and RL rewards. Human preference-based curation for question realism selects 100 tasks across 50 held-out worlds and 16 domains for DataWorld-Bench. The engine also yields 2,135 training tasks from 1,037 disjoint worlds. Evaluation of ten frontier models shows that the benchmark remains challenging, particularly for insight discovery, with a consistent computation-insight gap. RL-zero training on the synthesized tasks produces DataWorld-35B-A3B from Qwen3.6-35B-A3B without trajectory distillation. Under matched evaluation, it improves over its base by 7.06 points on DataWorld-Bench, 14.28 on DAComp, 20.90 on DABstep-Hard, and 12.13 on JobBench, a broader occupational benchmark.
Original Article
View Cached Full Text

Cached at: 10/16/26, 11:03 AM

Paper page - Data World: Scaling Data Environments from Scratch

Source: https://huggingface.co/papers/2610.21656 Authors:

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

Abstract

A multi-agent engine synthesizes data environments and tasks to train models for analytical insight discovery, improving performance across benchmarks without trajectory distillation.

Data agents use code to process heterogeneous files and derive analytical insights. Scaling their training faces costly environment curation, privacy and intellectual-property restrictions, and difficulty verifying open-ended findings without known generative truth. Our core insight is to specify the world behind the files before generating the files themselves. We introduce Data World, amulti-agent enginethat synthesizes environments, tasks, and rubrics from explicit mechanisms. Independent audits check target recoverability from observed files; rubrics support evaluation and RL rewards. Human preference-based curation for question realism selects 100 tasks across 50 held-out worlds and 16 domains forDataWorld-Bench. The engine also yields 2,135 training tasks from 1,037 disjoint worlds. Evaluation of ten frontier models shows that the benchmark remains challenging, particularly forinsight discovery, with a consistentcomputation-insight gap.RL-zero trainingon the synthesized tasks produces DataWorld-35B-A3B from Qwen3.6-35B-A3B withouttrajectory distillation. Under matched evaluation, it improves over its base by 7.06 points onDataWorld-Bench, 14.28 on DAComp, 20.90 on DABstep-Hard, and 12.13 on JobBench, a broader occupational benchmark.

View arXiv pageView PDFProject pageGitHub13Add to collection

Get this paper in your agent:

hf papers read 2610\.21656

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper1

#### Zhanli-Li/DataWorld-35B-A3B Reinforcement Learning• 35B• Updatedabout 3 hours ago • 47 • 1

Datasets citing this paper3

#### Zhanli-Li/DataWorld-Bench Updatedabout 3 hours ago • 133 • 1 #### Zhanli-Li/DataWorld-Train-2k-part2 RL Environment• Updated22 minutes ago • 75 • 1 #### Zhanli-Li/DataWorld-Train-2k-part1 RL Environment• Updatedabout 3 hours ago • 3 • 1

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2610.21656 in a Space README.md to link it from this page.

Collections including this paper1

Similar Articles

@omarsar0: Nice paper on scaling data environments from scratch. They built an engine that creates data-analysis environments from…

X AI KOLs Following

The paper introduces Data World, a multi-agent engine that synthesizes data-analysis environments, tasks, and grading rubrics from explicit generative mechanisms, enabling verifiable open-ended findings. RL-zero training on 2,135 generated tasks improves Qwen3.6-35B-A3B across multiple benchmarks, while evaluation of ten frontier models reveals a consistent gap between computation accuracy and insight derivation.

@Phoenixyin13: This latest blockbuster paper from Meta FAIR aims to tell the AI industry an important bellwether: "Large model data is ushering in the era of intelligent scientists." In this paper, a 4B small model precisely refined by Autodata not only crushes the same-scale models trained with traditional synthetic data on legal reasoning tasks, but also...

X AI KOLs Timeline

Meta FAIR's latest paper proposes the Autodata method, which uses an intelligent data scientist Agent to autonomously generate and optimize high-quality data, enabling a 4B small model to defeat a 397B large model on legal reasoning tasks. This indicates that data quality can bridge the gap in parameter count, providing new insights for data pipelines and scaling.