Hugging Face AI Updates: June 7, 2026

1. Build Small Hackathon Report Runs Five Small Models for Differentiated Agent Behavior

Hugging Face. A Build Small Hackathon field report published on the Hugging Face blog describes Thousand Token Wood v2, a game that drives five woodland-creature agents using four distinct small models from different labs, gpt-oss-20b, MiniCPM3-4B, Nemotron-Mini-4B, and a fine-tuned Qwen 0.5B, to produce genuinely differentiated behavior rather than relying on a single model. The author reports that the fine-tuned 0.5B model achieved 0 percent self-buys and 100 percent valid offers, outperforming its 3B teacher model. The writeup argues that serving-layer friction, addressed here with a tolerant JSON parse-and-repair layer and bounded sentiment-based relationship summaries, rather than modeling capability, is the main obstacle when combining heterogeneous small models. Source