Oc-e-Non-v1: a terse Qwen3-4B fine-tune

A Reddit user released Oc-e-Non-v1, a fine-tune of Qwen3-4B-Instruct-2507 that cuts mean response length by 50%, built with the Heretic tool. The author notes prompting could achieve the same and frames it as a light-hearted experiment while awaiting Qwen4.

• Heretic 2.0 Released as General-Purpose Model Modification Tool2026-11-01
• Heretic is on PewDiePie!2026-10-01
• Heretic Removes Restrictions From Language Models2026-09-21