Skip to content
← Models
Open weightsOpenBMB/10 trustMetrics-checked
MiniCPM-V-4.7-35B-A3B logo

MiniCPM-V-4.7-35B-A3B

OpenBMB's 35B MoE vision-language model — 3B active params, on-device speed, #2 on Hugging Face trending.

Visit official page →

About

MiniCPM-V-4.7-35B-A3B is OpenBMB's first mixture-of-experts entry in the MiniCPM-V series: 35B total parameters with only 3B active per token. It handles single images, multi-image inputs, and video understanding, and its headline claim is on-device MoE inference — 13+ tokens per second on an iPhone 17 Pro Max via a technique called Lookahead Expert Scheduling (provider-claimed). The weights are open under the Apache-2.0 lineage license. It sits at #2 on Hugging Face trending with 17.6k downloads, positioned as a new efficiency frontier for vision-language models that run on consumer hardware rather than datacenter GPUs.

Specs

Context window

Not disclosed on model card

Pricing

Free and open weights

Free quota

Full weights free to download from Hugging Face

API access

Self-host; community inference endpoints on Hugging Face Spaces

Weights

Open weights

Released

Not disclosed

Benchmarks

Positioned as a performance-efficiency Pareto frontier for on-device VLMs (provider-claimed); 13+ tok/s on iPhone 17 Pro Max. Independent benchmark replication pending.

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

vision-languageon-device-aiopen-weightsmobile-ai

Trust score — 5/10

/10 trustMetrics-checked

Scored 7/10 (METRICS tier, metrics-v1). Evidence: OpenBMB has a multi-year MiniCPM track record, the Apache-2.0 lineage license is genuinely permissive, and #2 on HF trending with 17.6k downloads shows real pull. Docked for: (1) the iPhone throughput and Pareto-frontier claims are provider-reported with no independent replication yet; (2) as a fresh release, long-term maintenance cadence is unproven; (3) benchmark coverage is thinner than established VLM families. License: Apache-2.0 lineage.

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
8.0/10
Maintenance · recency, releases, issue load
7.0/10
Openness · license + docs
9.0/10
Manual review · sanity read + red-flag check
7.0/10
  • · OpenBMB multi-year MiniCPM track record
  • · #2 on HF trending, 17.6k downloads
  • · Apache-2.0 lineage license is genuinely permissive
  • · manual review: iPhone throughput and Pareto-frontier claims are provider-reported, unreplicated

More in the minicpm-v family

Sources