| qwen2.5-coder:7b | 7B | Coding / agentic | EVO Arc | Q4_K_M | 2026-06-19 | evo-qwen25coder-7b | CLI | 9.4 | 223 @3558 | 2.3 | ~5GB | Fastest EVO coder. Always-on pick — leaves ~55GB free. |
| qwen2.5-coder:14b | 14B | Coding / agentic | EVO Arc | Q4_K_M | 2026-06-19 | evo-qwen25coder-14b | CLIMux | 5.5 | 112 @3558 | 10.1 | ~9GB | Balanced quality/speed. num_ctx 16K. |
| qwen2.5-coder:32b | 32B | Coding / heavy | EVO Arc | Q4_K_M | 2026-06-19 | evo-qwen25coder-32b | CLI | 2.7 | 53 @3558 | 23.7 | ~19GB | Best EVO quality, slow. On-demand only. |
| qwen3:14b | 14B | Coding (thinking) | EVO Arc | Q4_K_M | 2026-06-19 | evo-qwen3-14b | CLIMux | 5.4 | 111 @3545 | 8.8 | ~9GB | nothink template; repeat_penalty 1.05. |
| gemma3:12b | 12B | General + Vision | EVO Arc | Q4_K_M | 2026-06-19 | evo-gemma3-12b | CLI | 5.1 | 119 @3540 | 10.6 | ~8GB | Multimodal — also a vision model (free bonus). |
| qwen3.6:35b | 35B A3B | Coding / agentic | RTX 4090 | Q4 | 2026-06-19 | rtx-qwen36-35b | CLI | 41.6 | 1620 @3539 | 53.9 | ~20GB VRAM | Mux rtx4090 default. ~3B active/token. |
| qwen3-coder:30b | 30B A3B | Coding / agentic | RTX 4090 | Q4 | 2026-06-19 | rtx-qwen3coder-30b | CLI | 182 ⚡ | 7436 @3537 | 38.5 | ~18GB VRAM | A3B MoE blazes (3B active). 182 is short-run burst — sustained ~27. |
| llama3.2:3b | 3B | Small / edge | Jetson | Q4_K_M | 2026-06-19 | jet-llama32-3b | CLI | 3.8 | 280 @3554 | 14.7 | ~2GB | Tier-0 edge. |
| qwen2.5-coder:7b | 7B | Coding / edge | Jetson | Q4_K_M | 2026-06-19 | jet-qwen25coder-7b | CLI | ❌ failed | — | — | ~5GB | RemoteDisconnected — likely OOM/Frigate contention on the 8GB box. |