0.8B- Context
- 262Ktokens
- Released
- 1mo ago
- Size
- 35Bparameters
MiniCPM-V 4.7 35B-A3B is the newest generation of OpenBMB's MiniCPM-V series of multimodal vision-language models, built on a Qwen3.5-MoE base and the series' first sparse MoE variant. Its launch headline is on-device efficiency: the maker says it is the first open-source MoE model to run fluently on a phone, reaching 13+ tokens/s decoding on an iPhone 17 Pro Max. It is built for on-device image and video understanding.
- Multimodal model
- Open weights · 35B
- Accepts images, video
- Reasoning
- Tools
How it compares
Additional source evidence
- Source model
- openbmb/minicpm-v-4.7-35b-a3b
- Last updated
- 2026-10-07
- Model context
- 262K
- Model output limit
- 262K
- Open weights
- Yes
- File attachments
- Yes
- Reasoning
- Yes
- Tool calling
- Yes
- Temperature control
- Yes
Downloadable weights
Complete source record (JSON)
{
"id": "openbmb/minicpm-v-4.7-35b-a3b",
"name": "MiniCPM-V 4.7 35B-A3B",
"description": "First MoE MiniCPM-V model for single-image, multi-image and video understanding",
"last_updated": "2026-10-07",
"attachment": true,
"reasoning": true,
"tool_call": true,
"temperature": true,
"modalities": {
"input": [
"text",
"image",
"video"
],
"output": [
"text"
]
},
"limit": {
"context": 262144,
"output": 262144
},
"release_date": "2026-10-07",
"open_weights": true,
"license": "Apache-2.0",
"weights": [
{
"label": "Hugging Face",
"url": "https://huggingface.co/openbmb/MiniCPM-V-4.7-35B-A3B"
}
]
}Frequently asked questions
Running locally
MiniCPM-V 4.7 35B-A3B needs an estimated 24.9 GB at Q4, 43.3 GB at Q8 and 81 GB at FP16. These are estimates from the 35B parameter count, covering weights plus about 15% and 0.5 GB for a short context.
No. A 24 GB, 16 GB, 12 GB or 8 GB GPU does not fit MiniCPM-V 4.7 35B-A3B at Q4, Q8 or FP16 (estimate).
Yes, with enough unified memory. Assuming macOS lets the GPU use about 75% of unified memory, the smallest common Mac memory size that fits Q4 at an estimated 24.9 GB is 36 GB (estimate).
Privacy and data
Yes, you can self-host MiniCPM-V 4.7 35B-A3B for full control of your data because it has open weights.
Capabilities and specs
The context window of MiniCPM-V 4.7 35B-A3B is 262K tokens (262,144), and its maximum output is 262K tokens (262,144).
Images and video, yes. Audio and PDFs are not listed as supported. The input modalities of MiniCPM-V 4.7 35B-A3B are text, image and video.
Yes, MiniCPM-V 4.7 35B-A3B supports tool calling (function calling).
Yes, MiniCPM-V 4.7 35B-A3B supports a reasoning mode.
Quality and benchmarks
The closest alternatives, matched on model kind and release date with no comparable price, are LightOnOCR-3 0.8B by Lightonai, Flint by Recursive and Qwen3.8 Flash Next by Alibaba.
LightOnOCR-3 0.8BMiniCPM-V 4.7 35B-A3B vs LightOnOCR-3 0.8BFlintMiniCPM-V 4.7 35B-A3B vs FlintQwen3.8 Flash NextMiniCPM-V 4.7 35B-A3B vs Qwen3.8 Flash Next
Licensing and access
MiniCPM-V 4.7 35B-A3B is open weights, released under the Apache-2.0 license.
Yes. The Apache 2.0 license is a permissive license that allows commercial use.
You can download the MiniCPM-V 4.7 35B-A3B weights from the Hugging Face repository openbmb/MiniCPM-V-4.7-35B-A3B.
Basics
MiniCPM-V 4.7 35B-A3B was made by OpenBMB and released on October 7, 2026.
Yes. MiniCPM-V 4.7 35B-A3B is free to use as it has open weights.
Generated from Vertical Tabs data, updated November 11, 2026.
Discussion
No comments yet
Be the first to share your thoughts.