Model reference · open weights
moondream3.1 is an open-weight language model from moondream. moondream3.1-9B-A2B (FP8) weighs 10.5 GB; the smallest configuration that runs it is RTX 3060 12 GB.
For self-hosting, compare this build's requirements below with the GPU you plan to rent.
Summary of the moondream/moondream3.1-9B-A2B model card, 2026-10-01
What it is
| Released by | moondream |
|---|---|
| Released | 2026-06-30 |
| Parameters | 9.3B |
| VRAM | 10.5 GB for the weights |
What it runs on
How much memory each request adds isn't estimated yet for this architecture. The weights need at least the cards below, plus room for the context.
| Card | Weights alone |
|---|---|
| RTX 3060 12 GB FP8 without its speed-up here | tight |
| RTX 4060 Ti 16 GB | fits |
| RTX 3090 24 GB FP8 without its speed-up here | fits |
| RTX 4090 24 GB | fits |
| RTX 5090 32 GB | fits |
| L40S 48 GB | fits |
| A100 80 GB FP8 without its speed-up here | fits |
| H100 80 GB | fits |
| RTX PRO 6000 Blackwell 96 GB | fits |
| DGX Spark (GB10) 128 GB unified | fits |
| H200 141 GB | fits |
| B200 180 GB | fits |
From the model card
Moondream 3.1 is a vision language model with a mixture-of-experts architecture (9B total parameters, 2B active). It delivers state-of-the-art visual reasoning and detection while staying fast and cheap to deploy. Skills include query, detect, point, and caption, all native and all returning structured output.
For the full story on what's new — including how we trained it and how it holds up on your own tasks — see the release notes.
✨ Playground · ☁️ Cloud API · 📝 Release notes · 📄 License
Photon is Moondream's high-performance inference engine. It runs the model locally on NVIDIA GPUs (Ampere or newer) and Apple Silicon Macs, with the same API as Moondream Cloud.
pip install moondream
import moondream as md
from PIL import Image
# Runs locally via Photon. Weights download from Hugging Face on first use.
model = md.vl(local=True, model="moondream3.1-9B-A2B")
image = Image.open("photo.jpg")
# Caption
print(model.caption(image)["caption"])
# Visual question answering
print(model.query(image, "What's in this image?")["answer"])
# Open-vocabulary detection
for obj in model.detect(image, "person")["objects"]:
print(obj)
# Pointing
for point in model.point(image, "person")["points"]:
print(point)
No API key is required to run the base model locally. (You'll only need one to run finetunes or to call Moondream Cloud.) For supported hardware, streaming, finetunes, and production serving with Triton, see the local inference guide.
Prefer a hosted endpoint? The same skills are available through Moondream Cloud with a generous free tier. See the quickstart to make your first call.
Moondream 3.1 is released under the Moondream Model License.
Quoted from the model card on Hugging Face. The full card is behind the Hugging Face link above.
How it works