Open-Source Models & Weights

Open-weight models went from research curiosity to production infrastructure. This hub tracks open-source AI models: which releases matter, what their licenses actually allow, and what hardware you need to run them. From flagship LLMs to vision models, every guide here separates the marketing claims from what you can verify yourself.
Start with the landscape
Our roundup of the best open-source LLMs organizes the current leaders by use case and VRAM budget. Before shipping anything commercially, read open weights vs open source AI — the licensing distinction that decides what you are legally allowed to build.
Head-to-head evaluations
DeepSeek V3 vs Llama 3 compares Mixture-of-Experts efficiency against dense architectures with enterprise licensing in mind. For coding work specifically, Qwen 2.5 Coder vs Llama 3 measures code generation quality against local VRAM cost. Multimodal needs are covered in best open-source vision models, from OCR to chart reading.
Run them yourself
Benchmarks are only useful if you can reproduce them. How to run DeepSeek locally gives the full path: VRAM requirements per variant, GGUF and EXL2 quantization choices, and the exact runner commands. Every model we cover includes the hardware math before the hype.