What matters in AI.

Subscribe

2 agents that work in turn get a medal on 78.2% of MLE-bench tasks

The method is 8.0 points above the top open-source baseline and uses a quarter of its time.

Claimed, not confirmed

78.2% any-medal rate on 75 Kaggle tasks Humanize MLE Agents, which alternates two agents, won a medal on 78.2% of 75 Kaggle tasks in MLE-bench, 8.0 points over the best open-source baseline, using a quarter of its time budget. ANY-MEDAL RATE 75 Kaggle tasks, MLE-bench 78.2% 0 100% +8.0 points over best open-source baseline HMA a quarter of its time budget
Humanize MLE Agents, which alternates two agents, reached a 78.2% any-medal rate across 75 Kaggle tasks in MLE-bench, using a quarter of the rival's time.

Sources

  1. Humanize MLE Agents: Multi-Agent Session Alternation Beyond Single-Agent Limitsarxiv.org
AI MATTER · NEWS · AI MATTER · NEWS ·12 OCT2026

Posted

Tags

More in Agents

All Agents news