AIHacker Newsabout 2 hours ago
Show HN: I RL-trained an agent that trains models with RL (for, $1.3k)
1 min read
A developer built an RL system where a Qwen3.6-35B-A3B agent trains smaller models using GRPO. The outer loop runs 40 inner training jobs per batch across 16 GPU pods. The agent learned to select better base models and transfer skills to held-out tasks. Total cost for the headline arc was approximately $1,275.
Level
Hype check
Tap to vote and see what everyone thinks.
