Qwen/Qwen3.6-35B-A3B
Image-Text-to-Text • 36B • Updated • 6.12M • • 2.59k
Local-first LLMs
Small, capable models I run locally on a single RTX 3090 (Ollama / llama.cpp / transformers) — the backbone of self-hosted, sovereign AI.
Note Daily driver on the RTX 3090 (ollama qwen3.6:latest, Q4_K_M) - Qwen3.6 MoE, 35B total / ~3B active per token.
-
Qwen/Qwen3.6-27B
Image-Text-to-Text • 28B • Updated • 6.55M • • 2.11k
prism-ml/Ternary-Bonsai-27B-gguf
Text Generation • 4B • Updated • 698k • • 1.1kNote Qwen3.6-27B in ternary 2-bit — a 27B in 7 GB. Installed on the 3090 (Q2_0) next to its Q4_K_M sibling for quality A/B.
-
Qwen/Qwen3.5-9B
Image-Text-to-Text • 10B • Updated • 11.6M • • 1.77k -
Qwen/Qwen3.5-4B
Image-Text-to-Text • 5B • Updated • 6.33M • • 772 -
google/gemma-4-12B-it
Any-to-Any • 12B • Updated • 2.98M • 1.37k
unsloth/gemma-4-26B-A4B-it-qat-GGUF
Image-Text-to-Text • 25B • Updated • 417k • 347Note Official QAT GGUF of Gemma 4 26B MoE (4B active) — a 3090-friendly middleweight agent
-
Qwen/Qwen3-Coder-30B-A3B-Instruct
Text Generation • 31B • Updated • 1.54M • • 1.19k -
yuxinlu1/gemma-4-12B-coder-fable5-composer2.5-v1-GGUF
Text Generation • 12B • Updated • 245k • 2.78k -
yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF
Text Generation • 12B • Updated • 400k • 1.34k -
deepreinforce-ai/Ornith-1.0-9B
Text Generation • 1.47M • Updated • 2.02M • • 499
Qwen/Qwen3-VL-8B-Instruct
Image-Text-to-Text • 9B • Updated • 5.09M • • 1.02kNote Local vision on the RTX 3090 (runs as qwen3-vl:8b via ollama) — screenshots, docs, UI grounding
-
Qwen/Qwen3-Embedding-0.6B
Feature Extraction • 0.6B • Updated • 10.2M • • 1.13k