
Independent company · San Francisco, United States
MoondreamM87 Labs
Compact open-weight vision-language models for detection, captioning and segmentation, now part of Prime Intellect.
San Francisco, United States
- Founded
- 2024
- Type
- Independent company
- Based in
- San Francisco, United States
- Featured projects
- 4
Update Moondream announced on 14 October 2026 that it had been acquired by Prime Intellect, another lab in this atlas.
About
Moondream is a family of small vision-language models built by M87 Labs. The models handle captioning, querying, detection, pointing and segmentation, and run on GPUs, CPUs and edge hardware.
Moondream 3.1 uses a sparse mixture-of-experts design with 9 billion parameters, 2 billion of them active. The site reports more than five million monthly downloads, open weights and commercial use permitted. The company also offers a hosted API, a fine-tuning product and the Photon inference engine.
In October 2026 Moondream announced that it had been acquired by Prime Intellect, which says the combined team will accelerate open models and fast multimodal AI.
Mission and ambitionsMoondream aims to make capable visual AI small and fast enough to be fine-tuned and deployed on real-world edge cases anywhere.
Known for
- 01
Compact open vision-language models
- 02
Moondream 2 and 3 model lines
- 03
On-device and edge deployment
- 04
Photon inference engine
Milestones
- 2026
Moondream 3.1
A mixture-of-experts release with 9 billion parameters and 2 billion active, announced in July 2026.
- 2026
Photon inference engine
A megakernel inference stack that added speech models and NVIDIA GPU support through 2026.
- 2026
Parakeet Redux and Ultra
Speech-to-text models for local transcription in 25 languages, released in September 2026.
- 2026
Acquired by Prime Intellect
Moondream announced on 14 October 2026 that it was joining Prime Intellect.
Open source
Tiny vision language model
- moondream3-previewmodel
Moondream 3 preview
- moondream3.1-9B-A2Bmodel
Moondream 3.1 MoE
- megalith-mdqadataset
Image QA dataset
Notable releases
Openness
Apache-2.0 inference code; model weights and datasets on Hugging Face.
People
- Vik KorrapatiFounder
Joined Prime Intellect to lead inference work after the acquisition.
- Jay AllenFounder
Joined Prime Intellect alongside Korrapati.
Listed from the lab's own pages and public profiles.
Contact and community
Email printed on website.
Sources
Compiled from the lab's public pages, Hugging Face and GitHub. Last checked 2026-11-17. Something wrong or out of date? Tell us on Discord.



