HashMatrix retweeted
Why am I benchmarking RED (Xiaohongshu) for the new Wanderboat.ai GTM strategy proposal in the US? DATA speaks for itself:
🔹 350M+ MAU
🔹 1B+ daily searches
RED is no longer just a social platform; it’s a high-intent lifestyle search engine deeply rooted in the younger generation. It has the most successful O2O loop.
Wanderboat.ai has already powered millions of locals in the U.S. to explore new and trending restaurants, hotels, and more.
HashMatrix retweeted
Break the record: Not for seniors, but fueled by aging. How a youth tech product just conquered the world's most hyper-critical Japan.
Rokid’s @RokidGlobal @RokidJapan 636M+ Yen success has zero to do with selling gadgets to seniors. Young people don't buy tech for fun—they buy it to survive. 🧵
HashMatrix retweeted
China Tech trip, fully funded. 🇨🇳✈️
Who wants to see the REAL China this September?
My exclusive client is SPONSORING a select group of talented Tech influencers for an all-expenses-paid trip. Yes, flights and hotels—all on us.
🤖 Made with AI
HashMatrix retweeted
Why is China the only major economy that skipped the Email?
In our debut episode, @JFinnyc and I dismantle the myth of "efficiency" and talk about how WeChat changed the DNA of Chinese business.
Podcast🔍: Asian Bestie Daddy
podcasts.apple.com/us/podcas…
HashMatrix retweeted
The endgame of the HI + AI era: let humans do more truly human things.
AI should amplify human capability, not replace human judgment. The future may belong to a new partnership between HI and AI, where technology serves humanity and people remain at the center of decision-making.
#ChinaAMC #AI #Investing #Innovation #TechTrends #DeepTalk
Introducing Phoenix-4, the most advanced real-time human rendering model ever built.
Its the first real-time model to render every pixel at runtime, generate and control emotional states, listen actively, and actually behave the way humans do in conversation.
HashMatrix retweeted
OpenClaw + Contra = Agents buying creative services 24/7 🔥
For the first time AI agents can buy from freelancers autonomously.
This is an amazing development for everyone.
So I put together a doc with the top 20 services agents are buying right now.
Inside, it’ll show you:
→ Services
→ Price points
→ Package options
→ Profile positioning
Want to get the list?
Just RT + comment "AI" and I’ll send it over :)
You can also check out my partner Contra here → contra.com/
MiniCPM-o 4.5: Seeing, Listening, and Speaking — All at Once. 👁️👂🗣️
✨Beyond traditional turn-taking, we’ve built a Native Full-Duplex engine that allows a 9B model to see, listen, and speak in one concurrent, non-blocking stream.
Watch how it masters real-world complexity in real-time:
🔔 Proactive Auditory Interaction: Interrupts itself to alert you when it hears a "Ding!" while reading cards.
🎨 Temporal Flow Tracking: Follows your pen in real-time, narrating and "mind-reading" your drawing as you sketch.
🍎 Omni-Perception: Scans groceries & identifies prices on the fly.
✨Why it’s a category-leader:
📌Performance: Surpasses GPT-4o and Gemini 2.0 Pro on OpenCompass (Avg. 77.6).
📌Architecture: End-to-end fusion of SigLip2, Whisper, and CosyVoice2 on a Qwen3-8B base.
📌Efficiency: Full-duplex live streaming now runs locally on PCs via llama.cpp-omni.
The era of "Wait-and-Response" AI is over. Proactive, real-time intelligence is now open-source.
🚀Experience it on Hugging Face: 🔗huggingface.co/openbmb/MiniC…
#MiniCPM #Omnimodal #FullDuplex #EdgeAI #OpenSource #ComputerVision
🤔Think a 9B model has reached its limits? The radar evaluation suggests otherwise. 📈
MiniCPM-o 4.5 sets a new standard for efficient #MLLMs. As shown by the purple line, our 9B model achieves a "full-circle" performance that exceeds larger and specialized models alike.
⭐️ Breaking down the Radar Chart:
✅ Omni-Modal Live Streaming (Top): Dominates in real-time tasks like Daily-Omni and Video-Holmes, enabling fluid, proactive interaction.
✅ Speech Conversation (Right): Shows a massive leap over baseline models, outperforming specialized tools in VoiceBench and bilingual CER/WER metrics.
✅ Visual Understanding (Left/Bottom): Maintains elite-level vision with a 77.6 OpenCompass avg, matching Gemini 2.5 Flash on benchmarks like MathVista and AI2D.
✅ Balanced Efficiency: Proving that a well-architected 9B model can provide a more "complete" hexagonal profile than 30B+ or proprietary counterparts.
It’s not just about size—it’s about the synergy of vision, speech, and reasoning in one package.
Experience the power of 9B on #HuggingFace and help us grow with a ⭐ if you find it useful!
👇 Try it now: huggingface.co/openbmb/MiniC…
#MiniCPMo45 #OpenSourceAI #MLLM #EdgeAI
🥳 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
[BEST OF CES 2026]
CES 2026 is one of the biggest tech events of the year.
It is filled with announcements and product showcases from many renowned brands and startups around the world.
Now that they’ve wrapped it up, I have compiled the absolute craziest, coolest, or just plain interesting products revealed at the show into one thread.
Don’t forget to grab your snacks, and enjoy the thread!
GIVEAWAY LIVE 🌟 Meet Link 2 & 2C Pro...
✅ 1/1.3” large sensor for DSLR-like image quality
✅ Dual microphones with four audio pickup modes
✅ AI Tracking (Link 2 Pro) / Auto Framing (Link 2C Pro)
✅ Elgato Stream Deck integration for instant control
✅ The ultimate webcam for meetings, streaming, and more
WIN a Link 2 Pro for your setup!
1️⃣ Follow @insta360 and @insta360vc
2️⃣ Like this post and tag 2 friends. Easy!
Winners will be drawn on January 27. Good luck!
HashMatrix retweeted
We just flipped the scaling narrative: Agentic Density > Parameter Count.
#MiroThinker 1.5 operationalizes Interactive Scaling—agents that seek evidence, iterate, and revise in real time (with a time-sensitive sandbox to avoid hindsight leakage).
Result: 30B hitting frontier-class agentic search, ~$0.07/query (≈20× cheaper vs 1T-class baselines).
Fully open source, read more: research.miromind.ai/blog/in…
Try: dr.miromind.ai
GH:github.com/MiroMindAI/MiroTh…
HF: huggingface.co/miromind-ai
HashMatrix retweeted
Today we're announcing CATArena – A game-based benchmark for Code Agents, introduced in our latest paper.
Current benchmarks mainly assess end-to-end performance in static scenarios, which suffer from score saturation, and increasingly depend on expert annotation as agent capabilities improve.
CATArena asks models (GPT, Gemini, Claude, etc.) to write their own executable code in games like Holdem, Gomoku, Bridge, and Chess, compete with each others, and then they learn from each others' code.
We then run multi-round tournaments and turn them into benchmarks to evaluate how well models understand tasks, code, and learn autonomously in open-ended environments.
catarena.ai/replays
HashMatrix retweeted
🚀 LongCat-Image is now open-source!
A bilingual (CN/EN) image generation & editing model with exceptional consistency, clarity, and text precision.
🔄 Superior editing performance with accurate visual alignment
📝 Renders complex Chinese characters, incl. rare glyphs
🖼️ Realistic rendering powered by data strategy and RL training
⚡ Fast, low-VRAM, lightweight inference
⚙️ 6B hybrid DiT architecture achieves 20B+ MoE performance
📊 Objective benchmarks prove it! LongCat-Image leads open-source:
- Image Editing: ImgEdit 4.50 (open-source SOTA), GEdit CN/EN 7.60/7.64 rivals commercial ones;
- Text Rendering: ChineseWord 90.7 outclasses all competitors;
- Text-to-Image: GenEval 0.87 & DPG 86.8 is competitive with top-tier open-source & closed-source models.
We release multiple LongCat-Image versions (mid/post-training model) with the full training toolchain to support multilingual visual research.
📄 GitHub: github.com/meituan-longcat/L…
🤗 HF: huggingface.co/meituan-longc…
📑 Report: github.com/meituan-longcat/L…
HashMatrix retweeted
🔥 AMO-Bench Leaderboard update
@Google Gemini-3-Pro claims the crown at 63.1% (AVG@32) – new SOTA!
@Alibaba_Qwen Qwen3-Max-Thinking surges to 2nd with 57.4% (AVG@32), hot on its heels.
New insight: Although the current SOTA remains in closed-source models, the most valuable insight is open-source models’ remarkable pace of catching up. Meanwhile, at comparable performance levels, Gemini-3-Pro has significantly fewer average output tokens (more concise outputs), demonstrating superior output efficiency.
😊Open-Source:
[Huggingface]huggingface.co/datasets/meit…
[Github] github.com/meituan-longcat/A…
[Paper] arxiv.org/abs/2510.26768
[Project Page] amo-bench.github.io
🔥 Introducing AMO-Bench: Challenging the Boundary of LLM Reasoning on IMO-Level Problems
Original Problems: 50 brand-new problems designed by human experts
Guaranteed Difficulty: Cross-validated to ensure at least IMO-level difficulty
Automatic Grading: Hybrid grading algorithm with 99.2% scoring accuracy
Human-Annotated CoT: Facilitate further case analysis to improve model
Key Findings: @Kimi_Moonshot (🏆Kimi-k2-Thinking) achieved 56% SOTA performance, followed by GPT-5-thinking (high) and Qwen3-235B-A22B-Thinking-2507; The majority of results are below 40%.
Welcome to try AMO-Bench and join us in pushing the boundaries of reasoning capabilities in LLMs!
😊 Open-Source:
[Huggingface] huggingface.co/datasets/meit…
[Github] github.com/meituan-longcat/A…
[Paper] arxiv.org/abs/2510.26768
[Project Page] amo-bench.github.io/



































