Nous Research’s Arete-27B retrains Qwen’s politics with 53 million tokens
Nous Research fine-tuned Alibaba’s Qwen3.8-27B on about 53 million tokens and cut answers adopting Beijing’s official line from 22% to 0.9% on its…
Topic
Model evaluation
A cheaper model can approach a frontier judge—if it knows when to escalate
Hightouch reports that pairing a fast decision model with selective Opus review reached near-frontier accuracy at more than four times lower cost. The…