these guys made a small 9B Qwen 3.5 model 7x better than GPT-5.6 Luna at contract clause review, and it's 20x CHEAPER. exact clause quotes: > tuned Qwen 9B: 59.6% > GPT-5.6 Luna: 8.5% > same Qwen before tuning: 4.2% it's done using Overmind, a fine-tuning platform that trains small models on your own agents' data. AND it’s open source, make sure to check it out and star if you found it useful: https://github.com/overmind-core/overmin…