September 19, 2026
Qwen 3 plays Doom on a home laptop: 4 GB of VRAM was enough
Qwen 3 with 1,7 billion parameters in Q8 quantization plays VizDoom on a laptop with an RTX 3050 and 4 GB of video memory; the demo was posted on 19.09.2026. The model makes decisions through a custom wrapper atop llama.cpp: it extracts the model's internal signals and selects moves based on them.
Qwen 3 with 1,7 billion parameters in Q8 quantization plays VizDoom on a laptop with an RTX 3050 and 4 GB of video memory; the demo was posted on 19.09.2026. The model makes decisions through a custom wrapper atop llama.cpp: it extracts the model's internal signals and selects moves based on them.
The path leads to agents that live on home hardware, not in the cloud.
