Skip to content
← Models
Open weightsOpenBMB/10 trustVerified
MiniCPM-V 4.0 logo

MiniCPM-V 4.0

4.1B-parameter open vision-language model that runs on phones.

Visit official page →

About

MiniCPM-V 4.0 is OpenBMB's compact multimodal large language model, released March 27, 2026 with 4.1 billion parameters. It processes text, single images, multiple images, and video directly on-device, which makes it suited to offline applications where privacy and latency matter. OpenBMB positions its performance as comparable to models twice its size.

The weights are open under Apache 2.0 and the model is integrated into llama.cpp, so it runs on consumer hardware and even smartphones; OpenBMB ships an Android demo app that runs MiniCPM-V fully on-device via llama.cpp with no data leaving the phone. Later 4.x releases (4.5, 4.6, 4.7) continued the line with hybrid attention backbones.

Specs

Context window

Pricing

Free quota

API access

Weights

Open weights

Released

March 27, 2026

Benchmarks

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

Use cases

on-device-aiimage-understandingvideo-understandingedge-ai

Trust score — 8/10

/10 trustVerified

Scored 8/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 94; Trending score 80/100 from researcher assessment.

More in the minicpm-v family

Sources