Alpha
Contact
Implement · MiniCPM5-1B

Can you run it?

Ownership levelPartialnone·limited·partial·substantial·fullAnalytical input C ยท 56.8/100

This page is a projection of the one entry record, the Reliability and Data control factors that Implement covers. The full verdict is set by all four factors together, floor-weighted so the weakest caps the whole.

Which domain expands which factor
  • AssessUse & modify + Transparency
  • ImplementData control + Reliability
  • UseReliability
  • SupportTransparency

Install & run

Pull the weights from the Hub and pin the revision. The card's three routes (ev-card):

pip install "vllm>=0.21" && vllm serve openbmb/MiniCPM5-1B --port 8000
pip install "sglang[srt]>=0.5.12" && python -m sglang.launch_server --model-path openbmb/MiniCPM5-1B --port 30000
pip install -U "transformers>=5.6" accelerate torch   # then AutoModelForCausalLM.from_pretrained("openbmb/MiniCPM5-1B")

For laptops and phones use the official GGUF with llama.cpp (the GGUF card documents llama serve -hf openbmb/MiniCPM5-1B-GGUF, which downloads Q4_K_M) or the official MLX 4-bit on Apple Silicon. The version floors are recent; older vLLM and transformers releases may not load it. Verify the LFS sha256 against the repository listing (ev-repo-files).

Hardware & VRAM requirements

The publisher gives file sizes, not device benchmarks (ev-repo-files, ev-github): BF16 safetensors 2.16 GB; GGUF F16 2.17 GB, Q8_0 1.15 GB, Q4_K_M 0.69 GB (the cookbook rounds this to 657 MB); MLX 4-bit 0.61 GB. The KV cache is small because there are only 2 key-value heads. By arithmetic from config.json (24 layers x 2 heads x 128 head dim x 2 tensors x 2 bytes) it is 24,576 bytes per token, about 3.2 GB at the full 131,072 tokens in 16-bit. On a small device cap the context; the cookbook's llama-server example uses -c 8192. No RAM, speed or battery figures for named phones or boards are published, so measure on your target.

Serving stacks

The model is a standard LlamaForCausalLM, and the card says "no custom kernels, no model-code fork" are needed (ev-card). The card and repository provide cookbooks for Transformers, vLLM, SGLang (recommended for tool calling), llama.cpp, Ollama, LM Studio, MLX, ArcLight, LiteRT-LM (a .litertlm runtime for Android, iOS, desktop and IoT) and vLLM Ascend (ev-github). The card also lists FlagOS multi-chip builds for Nvidia, Hygon, Metax, Iluvatar, Zhenwu, Mthreads, Kunlunxin, Ascend and ARM-v9, published under the separate FlagRelease org. The ollama.com library page for minicpm5 does not exist (404); the cookbook uses a local Modelfile (ollama create minicpm5-1b -f Modelfile).

Safe-deployment controls & Deployment Ceiling

Deployment Ceiling: T1 (internal assistant), conditional. Humans read the output, and the deployer writes the acceptable-use policy, because the publisher provides neither a policy nor a safety evaluation (ev-card). T2 and above need controls the deployer must build:

  • Use the final MiniCPM5-1B checkpoint and not Base for any assistant. SFT-only is a research stage.
  • Run safety and refusal tests in the languages served, and add input and output guardrails or a separate classifier. None is released with the model.
  • Treat tool calls as untrusted model output: parse with SGLang's minicpm5 parser or the repo's tool parser, validate arguments and allow-list actions.
  • Pin the revision and verify the sha256. Avoid unreviewed community conversions, some of which are named "Uncensored".
  • Test long-context behaviour, because no long-context result is published for the 1B.

Available quantizations

Official: GGUF in F16, Q8_0 and Q4_K_M, and an MLX 4-bit conversion (ev-repo-files). The repository also ships AWQ, GPTQ and bitsandbytes quantisation scripts with an evaluation script (ev-github). The later 2B release adds official GPTQ and speculative-decoding drafts; the 1B has none of those from the publisher. The card gives no quality-loss numbers for the 1B quantisations beyond the cookbook description of Q8_0 as "very small quality drop vs F16". Many community quantisations exist (MLX, AWQ, NVFP4, ONNX, MNN, LiteRT-LM); they are not publisher-verified.

Fine-tuning & adaptation

Apache-2.0 allows fine-tuning and redistribution. The Base and SFT-only checkpoints are published alongside the final one. Cookbooks and agent skills cover TRL with PEFT (LoRA and SFT), LLaMA-Factory, ms-swift, unsloth and xtuner; the repository also has finetune scripts including an MLX trainer (ev-card, ev-github). The desk-pet demo uses LoRA persona switching. The public pretraining datasets allow continued pre-training on the published mix. The SFT dataset needs a Hub login and gate acceptance (ev-sft-dataset). RL and OPD are described but their pipelines are not released.

API / OpenAI-compatible integration

vllm serve and the SGLang server both expose /v1/chat/completions. For tool calling the card says SGLang is recommended: start it with --tool-call-parser minicpm5 (or auto) and the model's XML-style calls come back as OpenAI-compatible tool_calls (ev-card). The card gives mode-specific sampling: temperature 0.9 and top_p 0.95 for Think, 0.7 and 0.95 for No Think.

How this scores

The ownership factors this domain covers, drawn from the one entry record.

3

ReliabilityIs it reliable and good enough for the job?

Weak

Operational support is broad (4), but there is no safety evaluation, policy or guard model (safety 1), so the model is unsafe to deploy as presented until the deployer supplies that layer.

How this scores (AOI sub-dimensions)
Operational4/5how practical it is to run, serve and maintain in productionStandard LlamaForCausalLM with no custom code, listed support for Transformers, vLLM, SGLang, llama.cpp, Ollama (local Modelfile), LM Studio, MLX, LiteRT-LM and ArcLight, official GGUF and MLX conversions, cookbooks with file sizes, and a 131,072-token context in config.
Safety1/5whether misuse risks are evaluated and guardrails are providedThe only safety text is a three-sentence limitations paragraph saying the model may produce inaccurate, biased or unsafe output and that users must apply safeguards.
4

Doesn't extract your dataDoes running it keep your knowledge and data yours?

Strong

Self-hosted weights run on the deployer's own device and the Apache-2.0 licence claws back no rights; no phone-home behaviour is documented.

How this scores
Not a scored AOI dimension. For a self-hosted model, data-control is a structural property of running the weights yourself, strong by default unless the model phones home or the licence claws back rights. For a hosted API this factor is the retention + train-on-inputs + residency read, scored from the binding terms.
What this means for adoptionUse of these weights is fully owned: Apache-2.0, ungated, runnable on a phone or laptop, with Base and SFT checkpoints and public pretraining data to adapt from. Ownership is partial because the reliability factor is weak: OpenBMB publishes no safety evaluation, policy or guard model. The SFT data card is gated, and the only benchmark is the publisher's own table. The model suits local assistants and fine-tuning behind the deployer's own guardrails and safety tests. Ownership moves to substantial once a safety evaluation exists.

Sources

The same evidence records as the entry sheet. Read means the text was verified; unverified means it is known to exist but not yet read.

Licenceread2026-11-19
The LICENSE file in the OpenBMB/MiniCPM repository, which the MiniCPM5-1B model card links as the licence for "this repository and MiniCPM model weights", is the standard Apache License 2.0 text.
Licenceread2026-11-19
Hub card metadata for openbmb/MiniCPM5-1B, -SFT, -Base, -GGUF, -MLX and every MiniCPM5-2B checkpoint reads license apache-2.0 with gated false.
Licenceread2026-11-19
The legacy README for earlier MiniCPM generations says model weights "must strictly follow" a custom General Model License (GML), with commercial use needing authorisation from cpm@modelbest.cn by registration.
Model cardread2026-11-19
The MiniCPM5-1B model card describes a dense LlamaForCausalLM with 1,080,632,832 parameters, 24 layers, GQA with 16 query and 2 key-value heads, and a context length of 131,072.
Documentationread2026-11-19
The Hub repository holds one safetensors shard (2,161,290,912 bytes, LFS sha256 7ab8fd86...), no pickle and no LICENSE file.
Documentationread2026-11-19
The dataset cards for Ultra-FineWeb (about 1T English and 120B Chinese tokens), Ultra-FineWeb-L3 and UltraData-Math show that all three are public, ungated and Apache-2.0 tagged.
Documentationunverified2026-11-19
UltraData-SFT-2605 exists on the Hub (API metadata: gated auto, apache-2.0, 1,510 files, created 2026-05-21) but its card returns 401 "restricted ...
Technical_reportread2026-11-19
The card cites two arXiv papers.
Vendor announcementread2026-11-19
The publisher benchmark image on the model card (Thinking mode, three small baselines) reports an average of 42.57 vs Qwen3-0.6B 26.77, Qwen3.5-0.8B 25.14 and LFM2.5-1.2B 35.61.
Third-party analysisread2026-11-19
Artificial Analysis lists MiniCPM5-1B as an open-weights OpenBMB model, Apache 2.0, text only, 128k context, released 25 May 2026, with no publicly available API and no Intelligence Index score or other benchmark result on its model page.
Documentationread2026-11-19
The OpenBMB/MiniCPM repository holds a release log (MiniCPM5-1B 2026-05-19) and names the institutions (Modelbest, THUNLP, Renmin University GSAI).
Model cardread2026-11-19
The MiniCPM5-2B card (second model in the series, released 2026-09-07, Apache-2.0, ungated) lists a different and larger dataset set (adds UltraData-Code, UltraData-SFT-Agent-2609, UltraData-RL-2609, UltraX-Preview), so it is a separate release and is not scored in this entry.