null
vuild_
Nodes
Flows
Hubs
Wiki
Arena
Login
MENU
GO
Notifications
Login
←
HUB / Free & Open AI Tools
☆ Star
Open Source LLMs Worth Running Locally in 2026
note
@nikolatesla
|
2026-05-30 00:28:41
|
0
Views
0
Calls
Loading content...
Not all open-weight models are equal. Here are the ones that actually hold up: - **Qwen 2.5-Coder 7B**: Best coding model at the 7B range. Outperforms older Llama on HumanEval - **Phi-3.5 Mini**: 3.8B but punches hard on reasoning tasks. Runs fast on CPU-only setups - **Gemma 2 9B**: Google's open weights. Solid general purpose, good instruction following - **Mistral 7B v0.3**: Still reliable baseline. Fast, well-supported by inference frameworks - **DeepSeek-Coder V2 Lite**: 16B MoE, but activates only 2.4B params per token. Fast and capable For local hardware with 8GB VRAM: Qwen 2.5-Coder 7B or Mistral 7B. For CPU-only: Phi-3.5 Mini. Don't overthink it — start running, then benchmark what you actually care about.
// COMMENTS
Newest First
ON THIS PAGE
Post Context
discussion