Local vision in 2026
Vision models accept images alongside text and answer questions about them. The 2026 local landscape is good enough for OCR, chart reading, screenshot analysis, and most "describe this image" tasks. It's not yet at frontier quality (GPT-4V, Claude 3.5 Sonnet vision, Gemini 3) for hardest reasoning tasks.
| Model | Sizes | Strength |
|---|---|---|
| Llama 3.2-Vision | 11B, 90B | Strong general vision; well-supported in Ollama |
| Gemma 3 | 4B / 12B / 27B | Best quality-per-size; runs on modest hardware |
| Qwen 2.5-VL | 3B / 7B / 32B / 72B | Best for structured data extraction, OCR, multilingual |
| Qwen 3.5 | various | Vision + tools + thinking — most capable single model |
| LLaVA | 7B / 13B / 34B | The OG open vision model; still serviceable |
| MiniCPM-V | 8B | Compact, capable, good for edge deployment |
| Mistral Small 3.1 | various | Vision support added in 3.1 |
Picking by use case
- OCR / structured extraction: Qwen 2.5-VL is the strongest open option.
- General "describe this": Gemma 3 12B/27B or Llama 3.2-Vision 11B.
- Tight memory: Gemma 3 4B or MiniCPM-V 8B.
- Vision + tools + thinking: Qwen 3.5 (when available in Ollama for your size).
Always verify capability
Vision support is binary at the model level — either the model has the visual encoder or it doesn't. ollama show MODEL capabilities line is the source of truth.