Open-Weight AI Models to Run Locally

Nemotron is NVIDIA's family of AI models for developers building agents that reason, write code and use tools. Teams can evaluate and adapt the models using published weights, training data and training recipes.

CogView4 is a text-to-image model for image creators and developers who want to generate images on their own hardware. It accepts Chinese and English descriptions and can include Chinese text in the results.

Demucs separates mixed music recordings into parts for musicians who want to work with individual instruments or vocals. The author maintains a separate repository, with no active feature development and slow responses expected.

EmbeddingGemma is a local AI embedding model for developers adding search and retrieval to applications. It represents text, images, audio and video in a shared space so apps can find related content across media types.

Qwen3-Embedding is a family of models for developers building search and text analysis on their own hardware or servers. It represents text as numerical vectors so applications can compare passages by meaning.

PixArt-Sigma generates images from text prompts on your own machine or a self-hosted server. It's for developers and researchers who want downloadable image models and the code to train their own.

ModernBERT is a family of text encoder models for developers building search and classification systems on their own hardware. The base checkpoints predict masked words; retrieval, classification and question answering require task-specific fine-tuning.

CosyVoice is a text-to-speech system for developers building speech applications on their own hardware or self-hosted servers. It can generate speech in a reference speaker's voice, including in another language, without training a separate model for each speaker.

ColPali is a document retrieval library for developers and researchers building search or retrieval-augmented generation systems. It matches text queries to page images, using visual content alongside text. The colpali-engine package is deprecated, and maintainers recommend Sentence Transformers for new projects and production use.

Open-Sora is an AI video generation project for researchers, developers and creators who want to run or adapt a model on their own GPUs. It focuses on animating reference images with text guidance and also accepts text alone.

OpenCLIP is a Python library for developers and researchers building image-text matching and zero-shot classification applications. It supports pretrained CLIP models and training models on your own hardware.

Qwen-Image generates images from text and edits existing pictures. It's for creators and developers who want to run an image model on their own hardware, especially for work that needs readable text within images.

VoiceCraft generates speech from text and edits existing recordings using a few seconds of a speaker's voice as reference. It's for researchers and developers who want to work with voice generation on their own hardware without training a separate model for each speaker.

RWKV is a language model architecture for developers and researchers building LLM applications or training models. Its recurrent design keeps inference memory constant as text grows, making it relevant when a growing context cache would strain available memory.

WhisperSpeech is a text-to-speech system for developers and speech researchers who want to generate English audio with models they can run locally. It builds on OpenAI Whisper and provides pretrained models for use on your own hardware.

4M is a framework for researchers and developers who want to train and run models that combine images, text and other visual information. A single model can perform vision tasks and generate images from mixed or partial inputs.

StarCoder2 is a family of code generation models for developers who want code completion on their own hardware. These models predict continuations of code; they aren't instruction models, so conversational requests aren't their intended use.

Detectron2 is a Python library for developers and researchers building object detection and image segmentation applications. It provides model training tools and reusable components for computer vision research and production software.

Hunyuan-A13B is Tencent's language model for developers and researchers building self-hosted AI applications. It handles general text and reasoning tasks, with optimization for AI agent workloads.

ChatTTS generates conversational speech for developers and researchers building spoken LLM assistants or dialogue experiments. The released model is restricted to academic use and cannot be used commercially.

Granite is IBM's family of AI models for developers and businesses building applications on their own hardware or servers. You can customize the models for enterprise tasks, including coding assistants and document analysis.

StarDist detects individual cells and nuclei in crowded 2D microscopy images and 3D volumes. It's for researchers and bioimage analysts who need separate outlines for closely packed objects. Its shape-based approach suits objects that can be represented by star-convex polygons in 2D or polyhedra in 3D.

Distil-Whisper is a family of speech recognition models for developers who want to transcribe English audio on their own hardware. The models build on Whisper, reducing model size and processing time with some differences in accuracy.

SAM 3D Objects reconstructs objects in 3D from a single image. It's for researchers and developers who want to recover object geometry and appearance from photographs, including cluttered scenes with partly hidden objects.

Moondream is a vision language model for developers building apps that interpret images, such as warehouse or manufacturing footage. Its downloadable models can run on your own hardware, including offline in an air-gapped environment.

VoxelMorph is a Python library for researchers and developers who need to align medical images using learned deformations. It provides registration networks and training components, including unsupervised methods, with PyTorch and TensorFlow backends. The code is open source under the Apache 2.0 license.

Mochi 1 generates video from written prompts for creators and developers who want to run a video model on their own hardware or adapt it to their own footage.

Silero VAD is a pretrained voice activity detection model for developers who need to find speech segments in audio. It returns speech timestamps for use in audio processing pipelines.

FastSurfer processes structural human brain MRI scans on your own workstation or server, producing brain segmentations and cortical measurements compatible with FreeSurfer. It's for neuroimaging researchers working with individual scans or large cohorts who want shorter processing times within existing analysis workflows. The software is open source under Apache 2.0.

Mamba is a Python library for researchers and developers building sequence models, including language models. Its selective state space architecture processes sequences in linear time and is designed for efficient GPU execution.

MOSS TTSD generates spoken dialogue from scripts for podcasts, audio dramas, dubbing and commentary. Its focus is maintaining distinct voices across long conversations, including speaker changes and overlapping speech.

Indic TTS is a text-to-speech toolkit from AI4Bharat for developers and researchers building speech applications in Indian languages. It provides trained models and training code for generating speech on your own hardware.

mxbai-embed is Mixedbread's text embedding model family for developers building semantic search and retrieval systems. The downloadable mxbai-embed-large-v1 model converts queries and documents into vectors that applications can compare by meaning.

HunyuanVideo is an AI video generation model for creators and developers who want to generate clips on their own GPUs. Downloadable weights and inference code support generation from text prompts or an input image.