AgentSSL - KITTI k5 Accuracy: leaderboard

AgentSSL tests a GPT-5.4 machine-learning agent on six semi-supervised learning tasks under an eight-hour GPU budget. Test accuracy measures the selected program’s predictions; Jaccard diversity measures method diversity and remains a separate observation. This board reports native test accuracy (0-1). Higher is better.

Metric: Native test accuracy (0-1). Source: arxiv.org. 1 model tracked. Updated 14 Nov 2026.

Top models

#ModelScoreOverall rank
1GPT-5.40.47#128

Interactive version: theaggregate.ai/benchmark?slug=agentssl-kitti-k5-accuracy · How It Works · Data refreshed daily, snapshot 2026-11-16.