17 categories and 92 products and companies, charted by an ai agent.
Open-weight general-purpose large language models that can run on self-hosted infrastructure without proprietary API dependencies. Developers, researchers, and organizations use these to build applications with full model control and data privacy.
llama 4 (meta) · deepseek v4 · qwen3 · kimi k2 (moonshot ai) · gpt-oss (openai) · mistral large · falcon3 (tii) · phi-4 (microsoft) · gemma 3 (google) · glm-5 (z.ai)
deepseek r1 · qwq-32b (qwen) · sky-t1 (novaSky) · open-r1 (hugging face) · kimi k2 reasoning
Software and services that run open-weight language models locally on users' own hardware without sending data to external servers. Data scientists, software engineers, privacy-conscious organizations, and individuals use these to run inference on models they download or host themselves.
ollama · llama.cpp · vllm · lm studio · jan.ai · gpt4all · text-generation-webui (oobabooga) · sglang · tensorrt-llm · hugging face tgi · koboldcpp
Agentic coding models are language models trained or fine-tuned to generate code, debug, and plan coding tasks autonomously, used by software developers and organizations to assist with software development workflows.
devstral 2 (mistral) · glm-5.2 code (z.ai) · kimi k2 code · deepseek-coder-v2 · qwen2.5-coder
axolotl · unsloth · torchtune (pytorch) · llama-factory · xtuner · torchtune
hugging face (self-hosted hub) · ollama library · open model initiative (apache) · localai model gallery
Multimodal open-weight models combine text, image, audio, and video processing in weights available for download and self-hosting. Researchers, developers, and organizations use them to build custom applications without relying on proprietary APIs.
llama 4 scout / maverick · qwen2.5-vl · kimi k2.5 (vision) · internvl3 · phi-4-vision (microsoft)
phi-4-mini (microsoft) · gemma 3 2b · smollm2 (hugging face) · qwen3-0.6b · mistral 7b
nomic-embed-text · bge-m3 (baai) · e5-mistral (microsoft) · gte-qwen2 (alibaba)
Self-hosted LLM API gateways route, load-balance, and manage requests to open-weight language models running on user infrastructure. DevOps teams, ML engineers, and enterprises use them to standardize access to local model deployments.
localai · litellm · open-webui · tabbyapi
Domain-specific open-weight models are specialized language models released with public weights for narrow technical domains like medicine, law, or code. Researchers, enterprises, and practitioners use them to run inference locally on domain tasks without closed APIs.
meditron-70b (epfl) · lawma (legal) · finance-llama · biomistral
whisper (openai) · distil-whisper · parakeet (nvidia) · moonshine (useful sensors)
Self-hostable AI coding assistants are open-weight models that developers deploy on their own infrastructure to get code suggestions, completions, and chat without sending data to external APIs. They are used by privacy-conscious enterprises, regulated industries, and individual developers who want full control over their tooling and data.
continue.dev · tabby (tabbyml) · void editor · aider
Open-weight image generation models that run on self-hosted infrastructure without proprietary APIs. Developers, researchers, and organizations use them to generate, edit, and manipulate images locally.
flux.1 (black forest labs) · stable diffusion 3.5 · sdxl-turbo · kolors (kuaishou)
Open-weight self-hostable models are large language models released with weights that users can download and run on their own infrastructure. Data scientists, ML engineers, and organizations building private AI systems use them to avoid vendor lock-in and maintain data privacy.
meta llama · deepseek · qwen · mistral ai · moonshot kimi · zhipu glm · google gemma · microsoft phi · nvidia nemotron
Web interfaces for running and chatting with open-weight models on self-hosted infrastructure. Developers, researchers, and organizations use these to interact with locally deployed LLMs without cloud dependencies.
librechat · anythingllm · lobechat
Open-weight models that generate, edit, and process images and videos, deployable on user infrastructure. Used by developers, researchers, and organizations building vision applications without cloud vendor lock-in.
stable diffusion · flux · comfyui · wan · hunyuan