Open-source Models

Apache-licensed 2B–8B family with long context options; optimized for enterprise/local deployments.

Attention-free RNN-style open models with transformer-level performance and constant-memory inference.

Fully open training data, code, and checkpoints; transparent 7B/13B/32B family for reproducible research and apps.

Open Efficient Language Models (270M–3B) with code, weights, and training recipes for reproducible local use.

Microsoft’s 14B open-weight reasoning model yielding strong complex-task performance with modest hardware.

MosaicML’s Apache-licensed 7B family (base/instruct/long-context) widely used as a fine-tuning base.

Open-weight 32B RL-trained reasoning model reported to rival larger systems while remaining locally runnable.

Hugging Face repo for DBRX Instruct checkpoints under an open license for local inference and finetuning.

Family of open-weight models (1B–90B, some with vision) designed for local and edge deployment and strong instruction following.

01.AI’s upgraded Yi family (6B/34B) with improved instruction following and coding; widely used in local stacks.

Latest Qwen series from 0.5B–72B; dense decoder models with strong general, coding and math variants for local use.

Fully open multimodal (images/video+text) models & training stack; strong results and reproducible recipes.

Google’s open-weight 9B/27B models; practical to run locally and widely adopted for fine-tuning and apps.

Open multimodal family (1B–78B); 78B surpasses 70% on MMMU; broad image/video understanding.

Sparse MoE open model delivering top cost/performance among community LLMs; widely fine-tuned and quantized.