50+ model families supported. Here are the most popular ones — all tested, all ready to run locally.
| Model | Parameters | Quantization | Memory | Vision | Tools |
|---|---|---|---|---|---|
DeepSeek-V4 HOT DeepSeek | 685B (37B active) | Q4 | ~200 GB | — | ✓ |
Qwen3.6-35B-A3B HOT Qwen | 35B (3B active) | Q4 | ~20 GB | — | ✓ |
Gemma 4 HOT Gemma | 1B / 4B / 12B / 27B | Q4 / Q8 / FP16 | 1 GB – 16 GB | ✓ | — |
Qwen3 HOT Qwen | 0.6B – 235B | Q4 / Q8 / FP16 | 0.5 GB – 130 GB | — | ✓ |
Qwen2.5-VL HOT Qwen | 3B / 7B / 32B / 72B | Q4 / Q8 | 2 GB – 40 GB | ✓ | — |
Llama 4 Maverick HOT Llama | 400B (17B active, 128 experts) | Q4 | ~100 GB | ✓ | ✓ |
Mistral Small 3.1 HOT Mistral | 24B | Q4 / FP16 | ~14 GB | ✓ | ✓ |
Llama 3.3 Llama | 70B | Q4 / Q8 | ~40 GB | — | ✓ |
Llama 3.1 Llama | 8B / 70B / 405B | Q4 / Q8 / FP16 | 5 GB – 230 GB | — | ✓ |
Phi-4 Phi | 14B | Q4 / FP16 | ~8 GB | — | ✓ |
Command R+ Cohere | 104B | Q4 | ~60 GB | — | ✓ |
And 40+ more model families — Mistral, Mixtral, StarCoder2, Codestral, Nomic, Yi, and more.
GUI: Open NovaMLX → Models tab → Search or browse → Click Download.
CLI: nova download <model-name> — e.g. nova download Qwen3.6-35B-A3B-Q4