All tasks / Code & data / Coding
Best local model for coding on a 48 GB GPU
Open-weights models whose approximate minimum GPU memory is 48 GB or less, ranked from public benchmarks. Memory figures are approximate; quantised builds can need less.
CTop ranked: Qwen3.8-27B (Alibaba (Qwen)), 1,592 on LMArena WebDev / Code Arena (overall).10 options listed · data verified 19 Oct 2026
Ranked list
| Tier | Name | Approach | Runs | Rated by | Price | Skill |
|---|---|---|---|---|---|---|
| C |  Qwen3.8-27BAlibaba (Qwen) | Models | Local | 1,592 on LMArena WebDev / Code Arena (overall) | ~17 GB VRAM | — |
| C |  K2-Horizon-MoVA-36B-A4BMBZUAI Institute of Foundation Models | Models | Local | 25 on Artificial Analysis Intelligence Index | ~23 GB VRAM | — |
| C |  Nemotron 3.5 Bolt 74B-A9BNVIDIA | Models | Local | 21 on Artificial Analysis Intelligence Index | ~24 GB VRAM | — |
| C |  Gemma 4 (31B)Google DeepMind | Models | Local | 1,368 on LMArena WebDev / Code Arena (overall) | ~19 GB VRAM | — |
| C |  gpt-oss-20bOpenAI | Models | Local | 9 on Artificial Analysis Intelligence Index | ~16 GB VRAM | — |
| C |  Granite 4.2 30BIBM | Models | Local | 13 on Artificial Analysis Intelligence Index | ~18 GB VRAM | — |
| C |  Muse Glimmer 30BMeta | Models | Local | 1,355 on LMArena WebDev / Code Arena (overall) | ~19 GB VRAM | — |
| C | Nemotron 3.5 Lightning 30B-A3B NVIDIA | Models | Local | 13 on Artificial Analysis Intelligence Index | ~20 GB VRAM | — |
| C |  North Mini Code 1.0Cohere | Models | Local | 10 on Artificial Analysis Intelligence Index | ~19 GB VRAM | — |
| C | Qwen3-Coder-Next Alibaba (Qwen) | Models | Local | 9 on Artificial Analysis Intelligence Index | ~47 GB VRAM | — |
Tiers come only from public benchmarks (LMArena WebDev / Code Arena (overall), Artificial Analysis Intelligence Index); tools without a result of their own inherit the tier of the best model they offer. How we rank.