Models are generally chosen based on VRAM/RAM capacity. 7B-8B parameter models are ideal for 16GB RAM, while 30B+ requires 32GB+ RAM.
- Llama 4 (Meta): A state-of-the-art open-source model available in multiple sizes for advanced reasoning and coding.
- Qwen3 (Alibaba): Exceptional performance on coding and multilingual tasks, with specialized "Coder" and "Omni" (multimodal) variants.
- Mistral Large 3 & Small 3: High-performance, efficient models (Apache 2.0 license) optimized for speed and instruction-following.
- GPT-OSS (OpenAI): Open-weight models (20B-120B) providing GPT-4 level reasoning capabilities.
- DeepSeek R1/V3-Exp: Highly proficient in reasoning, math, and coding tasks.
- Gemma 3 (Google): Safety-focused, lightweight models (1B-27B) with strong performance on edge devices.
- Phi-4 (Microsoft): Small, highly efficient models (3.8B, 14B) designed for excellent reasoning on low-resource hardware.