Models
The open-weight model landscape, what fits which GPU, and licence limits.
| Page | Summary | Status |
|---|---|---|
| Landscape | The field has consolidated around MoE for big models and dense small models for local work. Qwen3.8-27B is the standout single-GPU dense model. | verified |
| What Fits Your VRAM | 8 GB → 7-8B dense; 16 GB → 12-14B; 24 GB → Qwen3.8-27B; 48 GB → 70B; 128 GB+ → server-class MoE. Q4 quantised. | verified |
| Licenses | Apache-2.0/MIT = genuinely open; Llama/Gemma/Qwen-Max/GLM-5.3/Kimi K3 = open-weight but restricted. License checks expire — verify before committing. | partial |