LongDS v1 - Social Good: leaderboard

Metric: Accuracy (%; macro-average over 10 social good tasks of per-task turn accuracy; DSGym ReAct agent in a persistent Jupyter kernel, up to 40 steps per turn, DeepSeek-V4-Pro judge). Source: arxiv.org. Saturation forecast: Around October 2027. 5 models tracked.

Top models

#ModelScore
1Gemini 3.1 Pro (Preview)41.73
2GPT-5.436.8
3Claude Sonnet 4.636.1
4DeepSeek V4 Pro32.41
5Kimi K2.631.29

Interactive version: theaggregate.ai/benchmark?slug=longds-v1-social-good · How It Works · Data refreshed daily, snapshot 2026-09-25.