Goekdeniz-Guelmez/JOSIE-1.1-4B-Thinking

JOSIE-1.1-4B-Thinking: Extended Context Reasoning Model

JOSIE-1.1-4B-Thinking, developed by Gökdeniz Gülmez, is a 4 billion parameter, full-weight fine-tuned model built on Qwen3-4B-Thinking. It is specifically optimized for deep reasoning capabilities across an extended context of 65,536 tokens. The model features uncensored outputs and a direct, honest personality, providing straightforward assistance without excessive agreeableness.

Key Capabilities & Features

• Extended Context: Supports a 65,536-token context length, enabling long-form reasoning and document analysis.
• Multilingual Support: Trained on Spanish, French, Portuguese, Italian, Arabic, Chinese, Japanese, Korean, Indonesian, Russian, Vietnamese, German, and Thai.
• Uncensored Output: Provides direct responses without built-in content filtering, requiring users to implement their own safety layers.
• Unique Training: Fine-tuned on a curated distillation dataset combining reasoning traces from Josie-Zero-8B and high-quality answer refinements from Anthropic Claude models (Sonnet 3.7/4.0, Opus 4.5/4.6).
• Apple Silicon Optimization: Trained using MLX frameworks on Apple Silicon, demonstrating viability on consumer hardware.

Ideal Use Cases

• Logical Reasoning: Complex multi-step deduction and problem decomposition.
• Mathematics & STEM: Problem-solving in algebra, calculus, statistics, scientific computing, and engineering.
• Creative Writing: Generating stories and dialogue with logical consistency.
• Document Analysis: Tasks requiring reasoning over extensive documents or multiple sources.