ImaJev-4b: I spent 15 days fine-tuning a 4B model to make business decisions from text and photos, and it just ranked #1 of 91 on JevBench & ahead of GPT-5.6 Luna on DecisionBench

A developer fine-tuned a 4B model, ImaJev-4b, over 15 days to make business decisions from text and photos. This model achieved the #1 rank out of 91 on JevBench (v1.4.2.2, scored 27 Sep) with a score of 67.37, surpassing Jev 1.13.0's score of 63.29. It also outperformed GPT-5.6 Luna on DecisionBench, demonstrating its advanced capabilities in business decision-making.

This report details how a 4B model, ImaJev-4b, not only ranked #1 out of 91 on JevBench but also surpassed GPT-5.6 Luna on DecisionBench, unlike other models of similar scale.