there's no one model that wins everything, so here's what i'd grab for each job. every name is linked to the exact build so you don't download the wrong one local text models - best overall: Huihui Qwen3.8 27B abliterated - agents: Huihui Qwen3.8 27B abliterated - fast: Qwen3.6 35B-A3B uncensored (HauhauCS) - coding: Qwen3-Coder 30B-A3B abliterated - writing: Gemma 4 26B-A4B ultra uncensored - roleplay: Cydonia 24B v4.3 - reasoning: Gemma 4 31B uncensored (Deckard thinking) - natural (no surgery): Hermes 4.3 36B - small GPU: Gemma 4 12B abliterated — Q4 is about 7GB - vision: Gemma 4 12B abliterated — grab the mmproj file too - laptop: Gemma 4 E2B uncensored — Q4 is about 3.4GB - biggest: GLM 4.6 derestricted v3 — 357B, not a home setup hosted (no GPU needed) - hosted: DeepSeek V4.1 Flash uncensored (Venice) - research: DeepSeek V4.1 Flash uncensored (Venice) — turn web search on - long context: GLM 5.3 Flash uncensored (Venice) — 1M tokens