MiniCPM Spark Program S1 concludes with 16 winning projects
The results show small models like MiniCPM5-2B and MiniCPM-o 4.5 running usefully on consumer hardware, from a Mac mini M4 to an ESP32-S3, without cloud reliance. Builders working on edge deployments can study the winning setups, open scripts and evaluation data on OpenBMB's GitHub and Hugging Face to replicate or extend them. The programme also signals OpenBMB using community challenges to grow its open-source ecosystem internationally.
OpenBMB's MiniCPM Spark Program S1 challenge has ended with 16 award-winning projects from developers in China, Spain, Italy, Australia and the US. Winners include a home camera assistant running on a Mac mini M4 that cut MiniCPM-V inference calls from about 720 per day to 60-70, and an ESP32-S3 flight radar using MiniCPM5-2B and VoxCPM2 for spoken tower-style announcements. Other projects feature a quantized MiniCPM-o 4.5 instance on an NVIDIA GB10 powering a robot with three roles, the Dubedo browser-based dubbing workspace, and Neko-OPSD's self-distillation research with MiniCPM5-2B, which released training data and 12400 evaluation responses. OpenBMB says the next season of the program is planned.
Small models, real-world impact. The MiniCPM Spark Program S1 challenge has wrapped up with 16 award-winning projects from developers across China, Spain, Italy, Australia, and the US! 🎉
Here’s a look at what the community built:
📷 A home camera assistant that cuts unnecessary inference Running locally on a Mac mini M4, a lightweight detector filters footage before MiniCPM-V analyzes relevant scenes and turns them into GIFs and daily reports. Model calls dropped from around 720 per day to just 60–70.
📡 A talking desktop radar An ESP32-S3 displays flight tracks and plays audio, while API calls to MiniCPM5-2B and VoxCPM2 provide tower-style announcements and speech. The demo uses simulated flight data. A small device, brought to life through hardware and model collaboration
🤖 One robot, three roles Powered by a single quantized MiniCPM-o 4.5 instance running on an NVIDIA GB10 system, a Reachy Mini Lite switches between companion, product guide, and performer without reloading model weights.
🎙️ Video dubbing you can keep refining Dubedo brings VoxCPM2 into a browser-based workspace where creators can edit dialogue, regenerate individual lines, preview audio, and export the mix.
🧪 Training experiments others can build on Neko-OPSD explores self-distillation with MiniCPM5-2B, sharing training data, methods, scripts, and 12400 evaluation responses across five experimental runs so others can inspect the results and build on the work.
Across applications, deployment tools, and training research, these projects show how much thoughtful engineering matters: choosing the right inputs, connecting components, testing limits, and sharing what actually worked.
Congratulations to all the winners, and thank you to everyone who contributed projects, code, and lessons learned!
Building with MiniCPM? Share your setup, model version, and real-world results with the OpenBMB community. Your experiment could help another developer take their next step.
S1 is complete. The exploration continues. See you next season!
👉 OpenBMB GitHub: https://t.co/TZa5EyDBCi 🤗 Hugging face: https://t.co/4zbRDKzkBh 👀 Official site: https://t.co/T04DpDUBLu
#MiniCPM #OpenBMB #SmallLanguageModels #OpenSource #EdgeAI
