The khazarai/Qwen3-4B-Qwen3.6-plus-Reasoning-Slerp is a 4 billion parameter language model with a 32768 token context length, developed by khazarai. This model is a surgical SLERP merge of two 4B reasoning models, specifically optimized for extreme logical reasoning, mathematical problem-solving, and precise Python code debugging. It achieves a "1+1=3 Synergy Effect" in Logical Inference and Planning, outperforming its base models and the official Qwen Thinking model on multi-domain reasoning benchmarks.
Loading preview...
khazarai/Qwen3-4B-Qwen3.6-plus-Reasoning-Slerp Overview
This model is a highly experimental 4 billion parameter language model developed by khazarai, created through a surgical SLERP merge of two powerful 4B reasoning models: khazarai/Qwen3-4B-Kimi2.5-Reasoning-Distilled and khazarai/Qwen3-4B-Qwen3.6-plus-Reasoning-Distilled. The primary goal was to combine deep analytical capabilities with mathematical and structural precision, resulting in a "1+1=3 Synergy Effect" in Logical Inference and Planning.
Key Differentiators & Performance
Unlike standard SLERP merges, this model employs a custom "Golden Path" (V5) strategy to prevent RAG degradation and maintain syntax adherence. This includes strictly pinning embed_tokens and lm_head to Qwen's vocabulary and applying a smooth gradient to intermediate attention and MLP layers. This approach has led to superior performance in multi-domain reasoning, scoring 77.18 on the khazarai/Multi-Domain-Reasoning-Benchmark, surpassing the official Qwen/Qwen3-4B-Thinking-2507 model (73.73).
Intended Use Cases
- Ideal for:
- Complex logical deductions
- Python code debugging
- Mathematical problem-solving
- Strict RAG (Retrieval-Augmented Generation) pipelines
- Not recommended for:
- Creative writing
- Poetry
- Highly imaginative storytelling
This model represents a trade-off, sacrificing some creative writing ability to maximize extreme logical reasoning and coding precision.