| Us | Model | Mida | Velocitat | Notes |
|---|
| Maxima qualitat | qwen3:32b | 18.8 GB | Lent | El millor general. RAM al limit. |
| Equilibri ⭐ | qwen3:14b | 8.6 GB | Mig | Recomanat dia a dia. |
| Rapid | qwen3:8b | 4.9 GB | Rapid | Conversa lleugera, respostes agils. |
| Lleuger | qwen3.5:4b | 3.2 GB | Rapid | Acceptable, molt agil. |
| Instant | qwen3:0.6b | 0.5 GB | Instant | Batch, scripts, automitzacions. |
| Us | Model | Mida | Velocitat | Notes |
|---|
| Frontier | qwen3-coder:30b | 17.3 GB | Lent | Millor coder. RAM al limit. |
| Qualitat alta | qwen2.5-coder:32b | 18.5 GB | Lent | Molt bo en HumanEval. RAM al limit. |
| Equilibri ⭐ | qwen2.5-coder:14b | 8.4 GB | Mig | Recomanat coding diari. |
| Rapid | qwen2.5-coder:7b | 4.4 GB | Rapid | Tasques simples, snippets. |
| Ultra lleuger | qwen2.5-coder:1.5b | 0.9 GB | Instant | Autocomplete basic. |
| Us | Model | Mida | Velocitat | Notes |
|---|
| Autocomplete FIM | codestral:22b | 11.7 GB | Mig | Fill-in-the-middle per editor. |
| Agent multi-fitxer ⭐ | devstral:24b | 13.4 GB | Mig | Agentic: debug, refactor, edits. |
| Us | Model | Mida | Velocitat | Notes |
|---|
| Chain-of-thought ⭐ | deepseek-r1:14b | 8.4 GB | Mig | Unic dedicat. Mates, logica. |
| Reasoning general | gpt-oss:20b | 12.9 GB | Mig | Complement per raonament complex. |
| Us | Model | Mida | Velocitat | Notes |
|---|
| Visio + tools ⭐ | gemma4:e4b | 9.0 GB | Mig | Imatges + tool calling natiu. |
| Agent lleuger | gemma4:e2b | 6.7 GB | Rapid | MoE eficient, automitzacions. |
| Us | Model | Mida | Velocitat | Notes |
|---|
| Batch / processar | qwen3:0.6b | 0.5 GB | Instant | 104 tok/s. El mes rapid. |
| Thinking mini | qwen3.5:0.8b | 1.0 GB | Instant | Thinking mode en <1 GB. |
| Pas intermedi | qwen3:1.7b | 1.3 GB | Rapid | Quan 0.6B no arriba. |
| Lleuger bo ⭐ | qwen3.5:2b | 2.6 GB | Rapid | Sorprenentment capac. |
Si nomes pots triar 4 models
General
qwen3:14b
Codi
qwen2.5-coder:14b
Reasoning
deepseek-r1:14b
Visio + Agents
gemma4:e4b