All tasks / Image / Image understanding
Best local model for image understanding on a 24 GB GPU
Open-weights models whose approximate minimum GPU memory is 24 GB or less, ranked from public benchmarks. Memory figures are approximate; quantised builds can need less.
ATop ranked: Gemma 4 (31B) (Google DeepMind), 1,262 on LMArena Vision Arena (overall, style control).4 options listed · data verified 19 Oct 2026
Ranked list
| Tier | Name | Approach | Runs | Rated by | Price | Skill |
|---|---|---|---|---|---|---|
| A |  Gemma 4 (31B)Google DeepMind | Models | Local | 1,262 on LMArena Vision Arena (overall, style control) | ~19 GB VRAM | — |
| A |  Qwen3.8-27BAlibaba (Qwen) | Models | Local | 1,244 on LMArena Vision Arena (overall, style control) | ~17 GB VRAM | — |
| B |  Muse Glimmer 30BMeta | Models | Local | 74.3% on Artificial Analysis MMMU-Pro (independent evaluation) | ~19 GB VRAM | — |
| C |  Molmo2-8BAllen Institute for AI (Ai2) | Models | Local | 1,112 on LMArena Vision Arena (overall, style control) | ~6.5 GB VRAM | — |
Tiers come only from public benchmarks (LMArena Vision Arena (overall, style control), Artificial Analysis MMMU-Pro (independent evaluation)); tools without a result of their own inherit the tier of the best model they offer. How we rank.