AI Agents for Data Analysis: The DSGym Framework
Together AI has released DSGym, a unified framework for training and evaluating LLM agents that perform data science tasks. It brings together 90+ bioinformatics tasks drawn from the scientific literature and 92 Kaggle competitions. The team trained a 4B model on synthetic data, and it achieves SOTA results among open-source solutions. The problem DSGym targets: existing benchmarks are incompatible with one another and don't require actual data analysis.
AI-processed from Together AI Blog; edited by Hamidun News
Together AI has built DSGym, a framework for training AI agents that tackle data analysis and bioinformatics tasks. The system combines 192 tasks from Kaggle and scientific datasets, so agents can train on real-world examples instead of synthetic ones.
Why do we need AI agents for data analysis?
Existing benchmarks for evaluating data science agents use incompatible formats and interfaces, and many of their tasks can be solved without any real data analysis. DSGym addresses this with a unified ecosystem of 192 real-world tasks.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.