Yuan Yao

Assistant Professor at College of AI, Tsinghua University. Lead MiniCPM-V and MiniCPM-o series models.

[AI's Nest @AINestHub1]

🚀 OpenBMB is one of the most exciting open-source AI research communities right now!

They are building the future of AI with powerful and efficient large language models that focus on speed, scalability, and real-world usability.

✨ Key features:

• Advanced LLM research & innovation
• Efficient model training (faster, lower cost)
• Lightweight AI models like MiniCPM
• Strong open-source ecosystem
• Tools for optimization, prompting & deployment

💡 What makes OpenBMB special is their focus on making powerful AI more accessible and practical for everyone, not just big tech companies.

A true game-changer in the open AI space! 🔥

👉 https://t.co/52TtGnJUH8⁠

@OpenBMB
#OpenBMB

[Shruti @heyshrutimishra]

China has a solution to every problem.

I needed to read signs, menus, and pharmacy labels.

This free AI app from OpenBMB runs entirely on my iPhone and does exactly that, in seconds, offline, in any language.

Bookmark this if you travel 🧵

[Madza 👨‍💻⚡ @madzadev]

In February 2026 @OpenBMB pushed MiniCPM-o 4.5 to GitHub, been reading through their technical report this week and the Omni-Flow architecture is genuinely interesting.

Omni-Flow treats input and output as parallel continuous streams. The model updates context mid-response, handles interruptions naturally, and can initiate without waiting for a trigger.

It has everything you need to actually work with it: complete and usable weights, full source on GitHub, a technical report that explains the design decisions, and a live demo.

Three months later the industry is treating the same capabilities as a breakthrough vision. That proves that the most unique releases in AI right now aren't always the loudest ones.

Stop watching AI. Start using it to build cool stuff.
#sponsored #ad

[OpenBMB @OpenBMB]

🚀MiniCPM-V 4.6 hits #1 on @huggingface Trending! 🏆
Huge thanks to the community for the incredible support!😘😘

🔗👇Try it:
🤗 Hugging Face: https://t.co/CEkwKMSBwc
💻 GitHub: https://t.co/iYDxpa52tn Modelscope:https://t.co/CHflKPLbvK
Web Demo: https://t.co/DYUrtD0YzM
App Demo: https://t.co/SL7IOhm6zv

What makes MiniCPM-V 4.6 stand out?

⚡ Ultimate On-Device Efficiency
Beats Gemma4-E2B-it and Qwen3.5-0.8B across key multimodal and Artificial Analysis benchmarks — scoring higher than Qwen3.5-0.8B using just 2.5% of its token budget.

👁️ Pro-Level Multimodal Intelligence
Exceptional fine-grained OCR, complex image reasoning, and multi-turn interaction in a highly compact footprint.

🛠️Built for developers
Fully open-sourced with out-of-the-box support for SGLang/vLLM/llama.cpp/Ollama, multi-platform mobile deployment, and low-barrier fine-tuning on consumer GPUs.

[Hunter Bown @goodhunt]

wait this is kinda crazy

https://t.co/ayr4ChEZN8

[Yuan Yao @yaoyuanAIR]

MiniCPM-V 4.6 is our most edge-deployment-friendly model to date. It uses LLaVA-UHD v4 architecture to boost visual encoding efficiency. Hope you like it! 🥰

[OpenBMB @OpenBMB]

1/5 MiniCPM-V 4.6 (1.3B) is now live 🚀🚀
High-res visual processing, optimized for consumer-grade and mobile hardware. We’ve leveraged the latest LLaVA-UHD v4 technique to cut vision encoding costs by 55%, enabling native edge deployment with extreme efficiency.
🔥 Beats Gemma4-E2B-it and Qwen3.5-0.8B across key multimodal and Artificial Analysis benchmarks — scoring higher than Qwen3.5-0.8B using just 2.5% of its token budget.
⚡ TTFT (75.7ms) 2.2x Faster than Qwen3.5-0.8B even with 3136² high-res images.
🏗️ ~1.5x Token Throughput compared with Qwen3.5-0.8B on a single RTX 4090.
Try the model here:
🤗 Hugging Face:
https://t.co/CEkwKMSBwc
💻 GitHub:
https://t.co/iYDxpa52tn
🔭 Modelscope:
https://t.co/CHflKPLbvK
🌐 Web Demo:
https://t.co/DYUrtD0YzM
📱 App Demo:
https://t.co/SL7IOhm6zv

[Adina Yakup @AdinaYakup]

MiniCPM V4.6 🔥 a 1B MLLM that actually runs on your phone, just released by @OpenBMB

✨ 1B - Apache2.0
✨ Runs on iOS, Android, HarmonyOS
✨ ~1.5× faster throughput than Qwen3.5 0.8B
✨ Mixed 4x/16x visual token compression

[Yuan Yao @yaoyuanAIR]

RT @lmsysorg: 🎉 Meet MiniCPM-V 4.6 from @OpenBMB, a 1.3B edge-friendly multimodal LLM with superior efficiency. Day-0 support is now live i…

[OpenBMB @OpenBMB]

1/5 MiniCPM-V 4.6 (1.3B) is now live 🚀🚀
High-res visual processing, optimized for consumer-grade and mobile hardware. We’ve leveraged the latest LLaVA-UHD v4 technique to cut vision encoding costs by 55%, enabling native edge deployment with extreme efficiency.
🔥 Beats Gemma4-E2B-it and Qwen3.5-0.8B across key multimodal and Artificial Analysis benchmarks — scoring higher than Qwen3.5-0.8B using just 2.5% of its token budget.
⚡ TTFT (75.7ms) 2.2x Faster than Qwen3.5-0.8B even with 3136² high-res images.
🏗️ ~1.5x Token Throughput compared with Qwen3.5-0.8B on a single RTX 4090.
Try the model here:
🤗 Hugging Face:
https://t.co/CEkwKMSBwc
💻 GitHub:
https://t.co/iYDxpa52tn
🔭 Modelscope:
https://t.co/CHflKPLbvK
🌐 Web Demo:
https://t.co/DYUrtD0YzM
📱 App Demo:
https://t.co/SL7IOhm6zv

[Yuan Yao @yaoyuanAIR]

Thanks to @_akhaliq for sharing our work! 🥳 MiniCPM-o 4.5 is our newest step toward more human-like, interactive multimodal LLMs. We'd love to hear your thoughts — try it out and let us know what you think! 💬✨

[AK @_akhaliq]

MiniCPM-o 4.5

Towards Real-Time Full-Duplex Omni-Modal Interaction

paper: https://t.co/wO3yzw0c2o

[OpenBMB @OpenBMB]

🚀 🚀Excited to announce the technical report of MiniCPM-o 4.5!

MiniCPM-o 4.5 transitions #AI interaction from traditional turn-based processing to a real-time, native full-duplex stream-based paradigm.

🌊 The Omni-Flow Framework
Instead of traditional VAD-based workarounds, we introduce the #Omni-#Flow framework. This unified stream paradigm aligns video, audio, and text on a synchronized millisecond timeline.
• Native Full-Duplex: Simultaneous perception and response.
• Proactive Interaction: Natively manages turn-taking without external VAD, supports proactive reminding.

📉 9B Scale, SOTA Performance
MiniCPM-o 4.5 demonstrates SOTA multimodal intelligence at its scale:
• Multimodal Benchmarks: Comparable to #Gemini 2.5 Flash on MMBench EN (87.6) and MathVista (80.1).
• Streaming Evaluation: 54.4% win rate on LiveSports-3K-CC, surpassing specialized models.

💻 The Ultimate Edge AI — Fully Functional without Network Connection
We are providing one-click installers for Windows (12G VRAM,RTX 5070) and macOS (M1-M5 Max/ M5 Pro).
• Local API Support: Deploy your own inference server to integrate native full-duplex into custom apps.
• Free Access: We are offering free community API services for exploration.
• 100% Private: Your data never leaves your machine.

Deploy in under 10 minutes. 🛠️👇

👐 Join the Open Future
The weights are open. The protocol is public.
📄 Technical Report: https://t.co/PjDakWcTbi
💻 GitHub: https://t.co/nI2IThWQSa
🤗 HuggingFace: https://t.co/KzzgiGXK5T
🌐 Web Demo: https://t.co/wqyCAPCcAy
#MiniCPMo #OpenSourceAI #EdgeAI #MachineLearning #ComputerVision #LLM

[JOY JULIET @JoyJuliet_AI]

This is what “native” multimodal should look like.

MiniCPM-o 4.5 moving to full-duplex with the Omni-Flow protocol feels like a clean break from stitched systems. Aligning video, audio, text, and speech on a shared timeline is a big architectural shift.

The 1Hz decision loop is especially interesting. The model actively chooses when to speak, listen, or interrupt, while still perceiving everything in real time. That’s much closer to human interaction.

Also impressive to see a 9B model hit 0.109 on OmniDocBench and lead in real-time duplex benchmarks. Efficiency is doing a lot of heavy lifting here.

No web search or tool calling yet, but as a reference design, Omni-Flow looks like something future systems will borrow from heavily.

[Markandey Sharma @TechByMarkandey]

Massive respect to the team 👏

The first open-source model with true human-like perception and interaction-seeing, hearing, and speaking in real time.

MiniCPM-o 4.5 raises the standard for omni-modal AI.

[OpenBMB @OpenBMB]

🥳 Introducing MiniCPM-o 4.5
The first full-duplex omni-modal LLM in open-source community 🎬🎙️

🔥 Key Highlights:
• Full-duplex Omni-modal Live Streaming: The model can see, listen, and speak simultaneously in a real-time conversation without mutual blocking
• Proactive Interaction: Moving beyond reactive QA to performing proactive interaction, such as initiating reminders
• Leading Performance: Scoring 77.6 on OpenCompass, it outperforms GPT-4o & Gemini 2.0 Pro in vision-language tasks with 9B params

The best part? You can experience all above on your PC!

#MiniCPM #OpenSource #MultimodalAI #LLM

[DailyPapers @HuggingPapers]

OpenBMB releases MiniCPM-V 4.5: An efficient MLLM powerhouse

This 8B parameter model achieves state-of-the-art visual reasoning, outperforming GPT-4o-latest and larger models with revolutionary efficiency.
Its 3D-Resampler enables high-FPS video understanding and robust OCR, even on your iPad.

[merve @mervenoyann]

I love MiniCPM-V 4.5, it's underrated

it's only 8B yet great in factual correction + thinking 💬

as they claim, gpt-4o level VLM on-device 👏 great work @OpenBMB

[Akshay 🚀 @akshay_pachaar]

GPT-4o level intelligence running on your phone!

MiniCPM-V 4.5 delivers enterprise-grade AI performance in just 8B parameters, outperforming models like GPT-4o, Gemini-2.0 Pro on vision and language tasks.

- 30+ language support
- Runs smoothly on iPhone/iPad

100% open-source!

[merve @mervenoyann]

MiniCPM-V 4.5 is very good! 🤗

it comes with hybrid thinking: it decides when to think on it's own 😍

it also can handle high res documents with odd aspect ratios, and super long videos efficiently 🙏🏻

see below hybrid results ⤵️ model is in comments!

[Adina Yakup @AdinaYakup]

MiniCPM-V 4.5 🚀 New MLLM for image, multi-image & video understanding, running even on your phone, released by @OpenBMB

https://t.co/MqAGmljhg0

✨ SOTA vision language capability
✨ 96× video token compression > high-FPS & long video reasoning
✨ Switchable fast vs deep thinking modes
✨ Strong OCR, document parsing, supports 30+ languages

[OpenBMB @OpenBMB]

🚀 Introducing MiniCPM-V 4.5 8B: pushing the boundary of multimodal AI!

～ SOTA VL Capability: Surpasses GPT-4o, Gemini 2.0 Pro, Qwen2.5-VL 72B on OpenCompass!
～ "Eagle Eye" Video: 96x visual token compression for high refresh rate and long video understanding
～ Controllable Hybrid Fast/Deep Thinking
～ Strong OCR & Doc Parsing: Surpasses GPT-4o & Gemini 2.5 on OmniDocBench

Get ready for the future of multimodal AI 👉
Huggingface｜https://t.co/JW7zPXOCqh
Github｜https://t.co/gtRJoHOlfd
Gradio｜http://101.126.42.235:30910/

#AI #MiniCM #GPT #Gemini #OpenBMB #ArtificialIntelligence #MachineLearning