# MiniCPM5-2B: Self-Hosting Hardware, Scenarios and Commercial License

> On-device and resource-constrained assistant baseline with official server, desktop and Apple-silicon runtime options.

- Verified: 2026-09-12
- Released: 2026-09-08
- Canonical: https://chinaapi.ai/open-model-deployment/minicpm5-2b/
- Back to directory: https://chinaapi.ai/open-model-deployment/
- Machine-readable dataset: https://chinaapi.ai/data/open-model-deployment.json

## Model facts

- Vendor: OpenBMB
- Parameters: 2.52B (1.98B non-embedding) total; 2.52B dense active
- Native precision: BF16; official GGUF and MLX variants
- Context: 128K tokens
- Category: General and agentic foundation models
- Modalities: text → text
- Serving paths: Transformers, vLLM, SGLang, llama.cpp, Ollama, LM Studio, MLX
- Official weights: Official BF16 / base: https://huggingface.co/openbmb/MiniCPM5-2B?utm_source=chinaapi&utm_medium=research&utm_campaign=open-model-deployment; Official GGUF: https://huggingface.co/openbmb/MiniCPM5-2B-GGUF?utm_source=chinaapi&utm_medium=research&utm_campaign=open-model-deployment; Official MLX: https://huggingface.co/openbmb/MiniCPM5-2B-MLX?utm_source=chinaapi&utm_medium=research&utm_campaign=open-model-deployment
- Model documentation: https://huggingface.co/openbmb/MiniCPM5-2B?utm_source=chinaapi&utm_medium=research&utm_campaign=open-model-deployment

## Deployment evidence

- **Official launch / minimum — evidence C:** Official llama.cpp, Ollama, LM Studio and MLX paths are published, but the vendor does not claim an absolute minimum RAM or VRAM configuration. Source: https://huggingface.co/openbmb/MiniCPM5-2B?utm_source=chinaapi&utm_medium=research&utm_campaign=open-model-deployment
- **200-person team — evidence E:** Benchmark a server runtime with the real output length and context mix; individual desktop compatibility is not a 200-seat capacity result.
- **Commercial API — evidence C:** Official vLLM and SGLang paths exist; add replicas, rate limits and quantization acceptance tests for public use. Source: https://huggingface.co/openbmb/MiniCPM5-2B?utm_source=chinaapi&utm_medium=research&utm_campaign=open-model-deployment

## Scenario evidence

- **On-device assistant — Vendor-stated:** OpenBMB positions MiniCPM5-2B for on-device and resource-constrained use and publishes GGUF and MLX variants. Source: https://huggingface.co/openbmb/MiniCPM5-2B?utm_source=chinaapi&utm_medium=research&utm_campaign=open-model-deployment
- **Local coding and tool use — Vendor-stated:** The card documents chat, reasoning and tool-use paths across mainstream local runtimes. Source: https://huggingface.co/openbmb/MiniCPM5-2B?utm_source=chinaapi&utm_medium=research&utm_campaign=open-model-deployment
- **Cross-platform private endpoint — ChinaAPI inference:** The broad runtime matrix makes the model a low-cost evaluation target before committing to larger checkpoints.

## Commercial-use check

- License: Apache-2.0 — https://huggingface.co/openbmb/MiniCPM5-2B/blob/main/LICENSE?utm_source=chinaapi&utm_medium=research&utm_campaign=open-model-deployment
- MaaS / hosted service: No model-specific MaaS restriction identified in Apache-2.0.
- Attribution: Provide the license and required notices, preserve attribution notices, and mark modified files; trademark rights are not granted.

## Avoid or validate first

- Assuming 131K context fits every official local format at useful speed
- Selecting a quantization without task-specific accuracy checks

## Known limitations and open questions

- No ChinaAPI reproduction or official absolute hardware minimum
- Long context materially increases KV-cache demand
- Quantized variants can change quality and throughput

## Evidence boundary

Loading weights, completing a first forward pass and meeting a production latency/SLA are separate thresholds. The 200-person and commercial tiers require workload-specific measurement. License summaries are product research, not legal advice.

Generated by GENERATED BY scripts/gen_open_model_deployment.py.