@heyshrutimishra: Full-sized AI models now run on phones. That's BitCPM, a new open-source model from ModelBest, Tsinghua, and OpenBMB. T…
Summary
BitCPM is a new open-source model from ModelBest, Tsinghua, and OpenBMB that uses ternary weights (-1,0,1) to run full-sized AI models on phones.
View Cached Full Text
Cached at: 05/25/26, 08:33 AM
Full-sized AI models now run on phones.
That’s BitCPM, a new open-source model from ModelBest, Tsinghua, and OpenBMB. The idea behind it is almost too simple: every weight is just one of three values, -1, 0, or +1.
They call it 1.58-bit, or ternary, and it’s what makes a model this capable small enough to live on your phone.
A model with weights this simple should be noticeably worse. BitCPM mostly isn’t.
Measured against the full-precision MiniCPM4 model at each size, it keeps 93-99% of the quality from 1B to 8B, and 90.1% at the smallest, 0.5B.
Roughly 6x lighter, with barely a dent in what it can do.
The other surprise is where BitCPM came from.
It wasn’t trained on the hardware most models are. The whole thing ran on Huawei’s Ascend 910B chips, start to finish, which makes it the first low-bit model proven end to end on that stack.
And it’s fully open. Weights and training scripts both, so anyone can rebuild it, not just run it.
A model this light stops needing the cloud.
It runs on your phone, your laptop, even the chip in your car, faster and more private because nothing has to leave the device.
BitCPM is open right now, the full family and the finetune scripts
Hugging Face： https://huggingface.openbmb.com/collections/openbmb/bitcpm4-cann…
ModelScope: https://modelscope.cn/collections/OpenBMB/BitCPM4-CANN…
If you build anywhere near edge AI, give it a real try, then tell the team what you find.
Similar Articles
@rohanpaul_ai: BitCPM-CANN just became the world’s first open-sourced 1.58-bit ternary LLM trained entirely on Chinese-developed AI in…
BitCPM-CANN is the first open-source 1.58-bit ternary LLM trained entirely on Chinese-developed AI infrastructure (Huawei Ascend 910B), offering extreme memory reduction for edge deployment.
@cjzafir: Fine-tune your first AI model today. Run GPT4o level model and run on your phone or laptop. @OpenBMB released 15M sampl…
OpenBMB released UltraData-SFT-2605, a 15M-sample high-quality SFT dataset for fine-tuning AI models like MiniCPM5-1B to run on phones or laptops.
@AdinaYakup: MiniCPM V4.6 a 1B MLLM that actually runs on your phone, just released by @OpenBMB 1B - Apache2.0 Runs on iOS, Android,…
OpenBMB has released MiniCPM V4.6, a 1B-parameter multimodal large language model optimized for mobile devices under the Apache 2.0 license. It features mixed visual token compression and claims approximately 1.5x faster throughput than Qwen3.5 0.8B while running natively on iOS, Android, and HarmonyOS.
@ModelScope2022: MiniCPM5-1B is now fully open source, including weights, training data, and deployment code. 1B params, #1 on Artificia…
MiniCPM5-1B is fully open-sourced with weights, training data, and deployment code; it achieves top scores among sub-2B models and runs on edge devices.
OpenBMB presents the model BitCPM-CANN 1.58 bit
OpenBMB introduced BitCPM-CANN, a 1.58-bit model being tested on Huawei Ascend 910B hardware.