OpenBMB launches MiniCPM5-2B-SFT, a 2B open-weights model for on-device text generation
OpenBMB released MiniCPM5-2B-SFT, a dense 2B Transformer for text generation, on Hugging Face under Apache-2.0. It is the second model in the MiniCPM5 series and targets on-device and resource-constrained deployment.
China context
- Original name
- 面壁智能
- Outside China
- Open weights · huggingface.co
- Claims
- Company-reported; not yet independently evaluated
- For builders
- Builders can download the model weights from Hugging Face and use the transformers library for local inference, enabling on-device AI applications without API costs.
- For investors
- The release of a 2B open-weights model with claimed SOTA performance in agentic tasks may pressure competitors in the small-model segment and signal continued investment in efficient model architectures.
OpenBMB released MiniCPM5-2B-SFT on Hugging Face. The model is a dense 2B Transformer for text generation, licensed under Apache-2.0 and built with the transformers library. It is the second model in the MiniCPM5 series, following MiniCPM5-1B, and is designed for on-device, local deployment, and resource-constrained scenarios. The release claims 2B-class open-source state-of-the-art performance, with advantages in coding, mathematics, long-context understanding, tool use, and agentic tasks compared to models of similar size.
MiniCPM5-2B-SFT is a dense 2B parameter Transformer model. The model card indicates it scales up the same training recipe as MiniCPM5-1B. It is released under Apache-2.0 and is compatible with the transformers library. The release includes a capability radar chart comparing MiniCPM5-2B with 4B-class models across dimensions such as code reasoning, math reasoning, instruction following, general knowledge, long context, tool use, coding agent, search agent, and general agent.
Developers outside China can now use a 2B-parameter open-weights model under Apache-2.0 for on-device and resource-constrained applications, reducing dependency on larger models for coding, math, and agentic tasks.
The Apache-2.0 license permits commercial use, enabling businesses to integrate MiniCPM5-2B-SFT into products without licensing fees, particularly for edge and mobile deployments.
Observable next signals include whether OpenBMB releases additional MiniCPM5 series models, whether the model is adopted in downstream open-source projects, and whether independent benchmarks confirm the claimed SOTA performance.