Key Features
β’ Precision Reasoning
-1. Optimized for accurate task execution and logical reasoning
-2. Excels in STEM tasks, scoring 80% on MATH and 75% on GPQA
-3. Outperforms similar-sized models in instruction-following
β’ Efficient Design
-1. 18 billion parameters, balancing power and resource use
-2. Runs on modest hardware (e.g., 16GB VRAM with 4-bit quantization)
-3. Low-latency inference for real-time applications
β’ Extended Context
-1. 64,000 token context window for detailed inputs
-2. Handles long-form documents and multi-step workflows
β’ Performance Highlights
-1. Competitive with Grok 2 and Phi-4 in reasoning and coding
-2. Strong at structured outputs, with 78% on HumanEval
β’ Developer-Friendly
-1. Open-source under MIT License, available on Hugging Face
-2. Supports function calling, APIs, and integration with tools like vLLM
β’ Versatile Applications
-1. Ideal for automation, research, and precision-driven workflows
-2. Powers chatbots, code assistants, and enterprise solutions
