· Models
Gemma
Google open multimodal models in two sizes, verified on H100.
Quickstart · pip install moondream
Call Gemma in a few lines.
python
# pip install moondream import moondream as md from PIL import Image # Or google/gemma-4-E4B-it for the larger variant model = md.photon("google/gemma-4-E2B-it") image = Image.open("photo.jpg") caption = model.caption(image)["caption"] print("Caption:", caption) answer = model.query(image, "What's in this image?")["answer"] print("Answer:", answer)
Full API reference · docs.moondream.ai/running-locally
| Model | Chip | Tested hardware | Since | Last verified | Benchmark |
|---|---|---|---|---|---|
| Gemma 4 E2B | B200 | NVIDIA B200 | v2.1 | Aug 30, 2026 | Results |
| Gemma 4 E2B | H100 | H100 SXM 80GB | v2.0 | Aug 3, 2026 | Results |
| Gemma 4 E4B | B200 | NVIDIA B200 | v2.1 | Aug 30, 2026 | Results |
| Gemma 4 E4B | H100 | H100 SXM 80GB | v2.0 | Aug 3, 2026 | Results |
Run Gemma with Photon
Install the SDK and pick any listed model.
$pip install moondream