#13

Qwen

Alibaba's open-weight LLM family spanning dense (0.6B, 32B) and MoE (up to 235B-A22B) architectures, trained on 36 trillion tokens across 119 languages. Covers text, vision-language (Qwen-VL), code (Qwen-Coder), math, audio, and agentic workloads. Post-training uses RLHF and DAPO; inference runs on vLLM, SGLang, and TensorRT-LLM. Most weights are Apache 2.0.