Ollama makes it simple to run LLMs locally. One-command setup for Llama 3, Mistral, Qwen, DeepSeek and 100+ models. Stars 100k+.
Leading open-source reasoning LLM from China. R1 model rivals OpenAI o1 in math and coding benchmarks while being completely free and open-weight.
Alibaba's open-source LLM family. Qwen 2.5 offers strong multilingual support, long context (128K), and competitive performance in a compact package.
High-throughput LLM inference engine with PagedAttention. Essential for production deployments needing low latency and efficient GPU memory management.
Meta's flagship open-source LLM family. Llama 4 brings multi-modal capabilities, longer context, and improved reasoning.
French AI lab behind the Mistral and Mixtral model families. Known for efficient open-weight models with strong performance.
Stability AI is the company behind Stable Diffusion, the open-source image generation model family. Also develops StableLM for language, Stable Code for programming, and Stable Video for video generation. Pioneered open-source gen
Google Gemma is an open-weight LLM family from Google available in 2B, 7B, and 27B parameter sizes. Offers state-of-the-art performance in a lightweight package for researchers and developers.
Microsoft Phi is a family of small, open-source language models (Phi-3, Phi-4) designed for efficiency and on-device deployment. Excels in reasoning, coding, and math despite compact size.
Nous Research creates state-of-the-art open-source fine-tuned language models including the Hermes series, Capybara, and Yarn-Mistral. Known for dataset quality, model alignment, and long-context breakthroughs.
Based on hedirbase's curated directory of open models, the most notable open source LLMs in 2026 include Ollama, DeepSeek, Qwen, vLLM, Meta Llama, Mistral AI, Stability AI, Google Gemma, Microsoft Phi, Nous Research. The right choice depends on your use case: frontier-class models for general reasoning, smaller dense models for self-hosting, and MoE models for high-throughput serving.
Open source LLMs are free to download and self-host under their respective licenses (Apache 2.0, MIT, Llama license, etc.), but you still pay for the compute to run them. Several of the listed models also offer hosted API versions for convenience.
hedirbase verifies the links and repo URLs of listed models on a regular basis and records a last-verified date on every entry page. This ranking reflects the current directory data; new models can be suggested through the Submit page at any time.