All tasks / Image / Image understanding
Best local AI for image understanding
Open-weights models and tools that run entirely on your own hardware, ranked from public benchmarks. Local tools are rated by the best open model they run.
STop ranked: GLM-5.3-Flash (Z.ai), 1,273 on LMArena Vision Arena (overall, style control).29 options listed · data verified 19 Oct 2026
Ranked list
| Tier | Name | Approach | Runs | Rated by | Price | Skill |
|---|---|---|---|---|---|---|
| S |  GLM-5.3-FlashZ.ai | Models | Local | 1,273 on LMArena Vision Arena (overall, style control) | ~185 GB VRAM | — |
| S |  OllamaOllama | Runtimes & gateways | Local | via GLM-5.3-Flash | Free | intermediate |
| A |  Gemma 4 (31B)Google DeepMind | Models | Local | 1,262 on LMArena Vision Arena (overall, style control) | ~19 GB VRAM | — |
| A |  DeepSeek-V4.1-Flash-1017DeepSeek | Models | Local | 1,260 on LMArena Vision Arena (overall, style control) | ~320 GB VRAM | — |
| A |  Step 5 PreviewStepFun | Models | Local | 1,250 on LMArena Vision Arena (overall, style control) | ~346 GB VRAM | — |
| A |  MiMo-V2.6-FlashXiaomi | Models | Local | 1,248 on LMArena Vision Arena (overall, style control) | ~178 GB VRAM | — |
| A | MiMo-V2.6-Pro Xiaomi | Models | Local | 1,246 on LMArena Vision Arena (overall, style control) | ~585 GB VRAM | — |
| A |  Qwen3.8-27BAlibaba (Qwen) | Models | Local | 1,244 on LMArena Vision Arena (overall, style control) | ~17 GB VRAM | — |
| A |  Muse Spark Open (816B-A42B)Meta | Models | Local | 80.9% on Artificial Analysis MMMU-Pro (independent evaluation) | ~470 GB VRAM | — |
| A |  Kimi K3Moonshot AI | Models | Local | 80.5% on Artificial Analysis MMMU-Pro (independent evaluation) | ~1600 GB VRAM | — |
| A |  MiniMax-M3MiniMax | Models | Local | 1,236 on LMArena Vision Arena (overall, style control) | ~245 GB VRAM | — |
| A |  Ling-3.0-flash-VLinclusionAI | Models | Local | 79% on Artificial Analysis MMMU-Pro (independent evaluation) | ~73 GB VRAM | — |
| A |  Open WebUIOpen WebUI Inc. | Apps | Local | via Gemma 4 (31B) | Free | intermediate |
| A |  llama.cppggml-org | Runtimes & gateways | Local | via Gemma 4 (31B) | Free | advanced |
| A |  vLLMvLLM project | Runtimes & gateways | Local | via Gemma 4 (31B) | Free | advanced |
| A |  LocalAIEttore Di Giacinto (mudler) | Runtimes & gateways | Local | via Gemma 4 (31B) | Free | advanced |
| A |  TextGen (text-generation-webui)oobabooga | Apps | Local | via Gemma 4 (31B) | Free | intermediate |
| A |  JanMenlo Research | Apps | Local | via Gemma 4 (31B) | Free | beginner |
| A |  SGLangSGLang project (LMSYS) | Runtimes & gateways | Local | via Gemma 4 (31B) | Free | advanced |
| A |  LMDeployInternLM (Shanghai AI Laboratory) | Runtimes & gateways | Local | via Gemma 4 (31B) | Free | advanced |
| A |  MLX-VLMBlaizzy (open-source project) | Runtimes & gateways | Local | via Gemma 4 (31B) | Free | advanced |
| A |  LM StudioElement Labs | Apps | Local | via Gemma 4 (31B) | Free | beginner |
| A |  Msty StudioMsty | Apps | Local | via Gemma 4 (31B) | Free | beginner |
| B |  Inkling SmallThinking Machines Lab | Models | Local | 1,208 on LMArena Vision Arena (overall, style control) | ~159 GB VRAM | — |
| B |  Mistral Medium 3.5Mistral AI | Models | Local | 1,200 on LMArena Vision Arena (overall, style control) | ~75 GB VRAM | — |
| B | Muse Glimmer 30B Meta | Models | Local | 74.3% on Artificial Analysis MMMU-Pro (independent evaluation) | ~19 GB VRAM | — |
| B | Inkling Thinking Machines Lab | Models | Local | 73.5% on Artificial Analysis MMMU-Pro (independent evaluation) | ~557 GB VRAM | — |
| B |  Llama 4 MaverickMeta | Models | Local | 1,145 on LMArena Vision Arena (overall, style control) | ~230 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.