AI Potluck
Model components / Base / pretrained models

MiniCPM

OpenBMB

OpenBMB's on-device model family, spanning MiniCPM3-4B through the MiniCPM4/4.1 8B hybrid-reasoning generation to the current MiniCPM5-1B. The line pairs small dense models with efficiency research — trainable sparse attention (InfLLM-V2), quantized end-side inference — to run capable models on phones and edge hardware.

The on-device counterpart to smollm on the map. Apache-2.0 across the line, with the efficiency research published as papers and some data artifacts (Ultra-FineWeb) but no end-to-end corpus or pipeline release, so open weights rather than open source. Adoption is dominated by MiniCPM5-1B and sits near the top of its band. Verified 2026-09-01 via the MiniCPM4.1-8B and MiniCPM5-1B HF records and the OpenBMB/MiniCPM repository.

Openness

3 high confidence
3.0
weights
open(Apache-2.0, ungated on HF across generations 3-5)
data
closed(training-data research published (Ultra-FineWeb, UltraClean) but the corpus is not released end to end)
code
partial(OpenBMB/MiniCPM is inference/deployment plus papers
license
Apache-2.0(OSI)

Apache-2.0 weights, ungated in every SKU read across generations 3 through 5, with training-data research published as papers and partial dataset artifacts but no end-to-end corpus release, and a repository that carries inference and deployment rather than the pretraining pipeline. Open weights, closed recipe: 3 by the fallthrough, governed by the current MiniCPM5 generation.

  • https://huggingface.co/api/models/openbmb/MiniCPM5-1B recorded 2026-09-01

    `gated: false`, cardData.license `apache-2.0`, created 2026-05-21; 810,294 downloads in the trailing 30 days — the current generation and the line's reach.

  • https://huggingface.co/openbmb/MiniCPM4.1-8B/raw/main/README.md recorded 2026-09-01

    Raw card: `license: apache-2.0`; links the GitHub repo and technical report (arXiv 2506.07900) and the InfLLM-V2 paper; a hybrid reasoning model card with no released training corpus and no pretraining pipeline link.

  • https://api.github.com/repos/OpenBMB/MiniCPM recorded 2026-09-01

    `license: Apache-2.0`, 10,288 stars, description "MiniCPM5-1B: A SOTA 1B on-device LLM, small yet powerful." — confirms MiniCPM5 is the current lead release; the repo is model usage/inference, not a training pipeline.

Adoption

3 high confidence
3.0

922,666 downloads in the trailing 30 days across the four declared artifacts (MiniCPM5-1B 810,294; MiniCPM4.1-8B 47,819; MiniCPM4-8B 37,939; MiniCPM3-4B 26,614), which bands at level 3 (100K-1M) on the model adoption scale.

Capability

3 medium confidence
3.0

The strong end of the small/on-device class: competitive with same-size Qwen3 and Gemma checkpoints and unusually efficient at long context via sparse attention, but the ceiling is the size class itself, well below the category's open-weight frontier. Level with smollm, the category's other small-model line, whose value records the same shape of claim (outperforms same-size peers, competitive one size up).

  • https://huggingface.co/openbmb/MiniCPM4.1-8B/raw/main/README.md recorded 2026-09-01

    The 4.1-8B card presents the line as "a hybrid reasoning model with trainable sparse attention" usable in deep-reasoning and non-reasoning modes, with benchmark comparisons against similar-size open models — the size-class-relative claim the band rests on.

Verified 2026-09-01