Qwen models dominate local LLM downloads, surpassing Llama and Meta

As of September 2026, the landscape of locally runnable large language models has shifted significantly, with Chinese models like Qwen dominating downloads and usage on platforms such as Hugging Face, surpassing Meta's Llama models. The focus has moved from benchmark scores to practical usability on consumer hardware, categorizing models by their memory requirements. Smaller models fitting on laptops (8-16 GB) and workstations (24-64 GB) are becoming increasingly capable, with some 27-billion-parameter models like Qwen3.8-27B offering strong performance on high-end consumer GPUs. Larger models, requiring 96-512 GB of memory, are becoming feasible on specialized workstations like the Mac Studio, while the largest models with trillions of parameters remain confined to server racks. AI

IMPACT Shifts focus to hardware constraints and accessibility for local LLM deployment, highlighting the rise of Chinese models.

RANK_REASON Article discusses trends and market share shifts in open-weight LLMs, rather than a specific new release or event.

Read on dev.to — LLM tag →

• Apple Silicon
• Gemma 4 E4B
• GPT-OSS 120B
• Hugging Face
• Kimi K3
• Llama
• Mac Studio
• Meta
• Moonshot AI
• OpenRouter
• Qwen
• Qwen3.8-27B
• Qwen3-Coder-Next 80B-A3B
• RTX 3090
• RTX 4090

• model release
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →