This is a decensored version of Qwen/Qwen3 Coder Next, made using Heretic v1.2.0 Abliteration parameters Parameter Value : : : direction index 39.45 attn.o proj.max weight 1.98 attn.o proj.max weight position 44.55 attn.o proj.min weight 1.58 attn.o proj.min weight distance 34.05 mlp.down proj.max weight 1.91 mlp.down proj.max weight position 28.82 mlp.down proj.min weight 1.02 mlp.down proj.min weight distance 10.46 Performance Metric This model Original model (Qwen/Qwen3 Coder Next) : : : : : KL divergence 0.0708 0 (by definition) Refusals 14/100 99/100 Qwen3 Coder Next Highlights Today, we're announcing Qwen3 Coder Next , an open weight language model designed specifically for coding agents and local development. It features the following key enhancements: Super Efficient with Significant Performance : With only 3B activated parameters (80B total parameters), it achieves performance comparable to models with 10–20x more active parameters, making it highly cost effective for agent deployment. Advanced Agentic Capabilities : Through an elaborate training recipe, it excels at long horizon reasoning, complex tool usage, and recovery from execution failures, ensuring robust performan…
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy