Data World generates data-analysis RL environments, reports gains on three benchmarks

Data World, a new engine, generates data-analysis environments from scratch by first defining the mechanisms behind each dataset, then producing files, tasks, and grading rubrics.

According to a paper summarized by Elvis Saravia (@omarsar0), training Qwen3.6-35B-A3B with reinforcement learning on 2,135 of these tasks, without distilled trajectories, adds 20.9 points on DABstep-Hard, 14.28 on DAComp, and 12.13 on JobBench.

The paper also reports that ten frontier models compute correct numbers more often than they draw correct conclusions from them.