MiniCPM-V 4.6 is a multimodal foundation model developed by OpenBMB. With a parameter scale of approximately 13.0B, it supports a long context window of 262K.
Parameters
13.0B
Context Window
262K
License
Apache 2.0
Release Date
2026-05-11
API Pricing
API pricing for this model is not yet available
Strengths
- ・High context processing capability
- ・Efficient model size
- ・Open license model
Weaknesses
- ・Limited knowledge vs. massive models
- ・Insufficient validation for specific tasks
- ・Variable hardware requirements
Use Cases
- ・Long-document analysis
- ・Multimodal information processing
- ・AI implementation on edge devices
Deep Analysis
Architecture
VLM (1.3B total)
SigLIP2-400M + Qwen3.5-0.8B
Context Window
262K tokens
~393 A4 pages
Release Date
May 11, 2026
License
Apache 2.0
Modalities
Text + Image + Video
Up to 128 frames
Intelligence Index
13 (Artificial Analysis)
#3 of 28 models in size class
Platforms
iOS, Android, HarmonyOS
Phone-optimized
Strengths
- ・Smallest MiniCPM-V ever at 1.3B parameters
- ・262K context window
- ・Runs on phones (iOS/Android/HarmonyOS)
- ・Apache 2.0 license
- ・Outperforms Qwen3.5-0.8B at 19x lower cost
- ・50%+ FLOPs reduction in visual encoder
- ・Supports vLLM, SGLang, llama.cpp, Ollama
Weaknesses
- ・1.3B limits complex reasoning
- ・Phone-optimized may sacrifice some quality
- ・Very new model (May 2026)
Competitor Comparison
| Model | Arena | SWE | GPQA | Price |
|---|---|---|---|---|
| Qwen3.5-0.8B | - | - | - | Free (open) |
| MiniCPM-V 4.0 | - | - | - | Free (open) |
| InternVL2-2B | - | - | - | Free (open) |
MiniCPM-V 4.6 is OpenBMB's 1.3B multimodal model built for phones. Released May 2026, it offers 262K context, image/video understanding, and runs on iOS/Android/HarmonyOS. It outperforms Qwen3.5-0.8B at 19x lower token cost.
Analysis generated: 2026-05-24