Browse practical GGUF models for local chat, vision and image-generation workflows — from compact 0.5B models to large 120B-class deployments.
models.realignsinc.com is a curated catalog. Download buttons point directly to original Hugging Face model files or repositories; Realigns Inc. does not re-host third-party weights. RAM figures are practical planning ranges, not guarantees — context size, runtime, GPU offload and concurrent workloads change memory use.
Compact models for 4–8 GB and entry-level local workstations, plus RINC's own 0.5B repository.
Lightweight local chat, business assistance and compact offline deployments.
Realigns model repository. Select the quantization file that matches your runtime and hardware.
General chat, multilingual work, drafting and compact local assistance.
Direct Q4_K_M download.
General assistant, summarization and local productivity.
Approx. 2.02 GB model file.
Compact reasoning, code assistance and local productivity.
Approx. 2.39 GB model file.
Text plus image understanding with a compatible multimodal runtime.
Approx. 2.49 GB model file; multimodal use can require projector assets.
Use the model size, quantization and context length that fit your actual machine. Larger is not automatically better for every workflow.
| Model | Brand | Size | Quant | Planning RAM | Use | Links |
|---|---|---|---|---|---|---|
| Qwen2.5 3B Instruct | Alibaba Qwen | 3B | Q4_K_M | 6–8 GB | Compact general assistant | DownloadModel Page |
| Llama 3.2 3B Instruct | Meta | 3B | Q4_K_M | 6–8 GB | General-purpose local assistant | DownloadModel Page |
| Phi-3.5 Mini Instruct | Microsoft | 3.8B | Q4_K_M | 6–8 GB | Compact reasoning and coding | DownloadModel Page |
| Gemma 3 4B IT | 4B | Q4_K_M | 8 GB+ | Text and vision capable family | DownloadModel Page | |
| Mistral 7B Instruct v0.3 | Mistral AI | 7B | Q4_K_M | 8–12 GB | Balanced local assistant | DownloadModel Page |
| Gemma 3 12B IT | 12B | Q4_K_M | 16 GB+ | Higher-quality text / multimodal workflows | DownloadModel Page | |
| Qwen2.5 14B Instruct | Alibaba Qwen | 14B | Q4_K_M | 16 GB+ | Stronger reasoning and business workflows | Model Page |
| Qwen2.5 32B Instruct | Alibaba Qwen | 32B | Q4_K_M | 32 GB+ | Large local reasoning model | DownloadModel Page |
| GPT-OSS 120B | OpenAI | 120B MoE | Q4_K_M | 96 GB+ planning range | Very large local model; repository uses split files for large quantizations | Model Page |
VLMs analyze images and produce text. They do not replace a dedicated image generator.
Lightweight image understanding and multimodal assistant workflows.
Use a compatible llama.cpp build; multimodal models may also require projector files.
Compact OCR, image understanding and document/visual analysis.
Main Q4_K_M model file; projector assets may be required by your runtime.
Image-to-text, question answering, summarization and visual understanding.
Multimodal family; check the repository for the matching projector file.
Stronger visual reasoning, OCR and complex image/document analysis.
Use the Q4_K_M variant and matching multimodal projector supported by your runtime.
Higher-quality multimodal understanding and enterprise knowledge workflows.
Main model file only; check the model repository for multimodal projector requirements.
These are GGUF image-model conversions. Image generation usually requires a compatible image runtime plus supporting text encoders and VAE assets.
Fast high-quality generation
Core diffusion/transformer file. ComfyUI-GGUF or another compatible runtime may require text encoders and VAE separately.
Detailed photorealistic and design workflows
FLUX.1-dev license restrictions apply. Additional runtime assets are normally required.
Higher-precision local image generation
Choose Q8_0 from the repository when hardware allows. Image runtimes typically need supporting encoders/VAE.
High-quality text-to-image generation
Core model file; compatible runtime and supporting assets are required.
Maximum local quantized precision in this catalog
Large memory requirement. Verify your image runtime and license before download.
Model names and trademarks belong to their respective owners. Inclusion in this library is informational and does not imply endorsement, partnership or ownership. Always verify the original Hugging Face repository before use.
