ModelRefs / Best Open-Source AI Models in 2026
Best Open-Source AI Models in 2026
Open-weight AI models you can self-host: Llama 3, Stable Diffusion 3, Whisper, BGE-M3 and more. Licenses, hardware and deployment guides.
Overview
Open-source models give you full control over weights, fine-tuning and data privacy. Browse the leading open models across text, vision and audio.
ModelRefs tracks 20 open source models with a published reference page. Each page states the provider, the capabilities ModelRefs has evidence for, the benchmark results it holds with their dates and sources, and the limitations and coverage gaps that remain.
Models in this category
20 open source models have a published reference page on ModelRefs.
Listed alphabetically. Category membership means a model is a candidate worth evaluating for this kind of work, not a ranking and not an endorsement — compare the evidence on each page against your own requirements.
- BGE-M3
- Command R+
- DeepSeek R1
- DeepSeek V3
- Gemma 3 27B
- InternVL2.5 78B
- Llama 3.1 405B
- Llama 3.1 70B
- Llama 3.1 8B
- Llama 4 Scout
- Mistral Nemo
- Mistral Small 3.1
- Mixtral 8x7B
- Nemotron-4 340B
- Phi-4
- Pixtral 12B
- Qwen 2.5 72B
- Qwen 2.5 Coder 32B
- Qwen2.5-VL 72B
- QwQ 32B
For an ordered view of the same ground, see the open source models ranking, which scores models on the benchmark evidence ModelRefs holds.
Other model categories
Continue your research
Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to Best Open-Source AI Models in 2026.
Frequently asked questions
What does open-source mean for AI models?
It usually means the model weights are downloadable and runnable locally. Licenses vary — some allow commercial use, others restrict it.
Can I run open-source models locally?
Yes. With quantisation, models like Llama 3 8B run on consumer GPUs and even modern laptops. Larger 70B+ models need multiple high-end GPUs.