12-24-Daily AI News Daily

AI Daily News 2025/12/24

AI News | Daily Briefing | Web-wide Data Aggregation | Cutting-edge Science Exploration | Industry Voice | Open Source Innovation | AI & Humanity's Future | Visit Web Version | Join Group Chat

Today’s Digest

Wenxin 5.0 Preview tops the domestic charts.
DingTalk launches the world's first AI Work OS.
MiniMax releases the strongest open-source coding model.
Zhipu AI officially unveils the GLM-4.7 flagship.
Alibaba's RTPurbo boosts inference speed by fivefold.

Product & Feature Updates

1. Wenxin 5.0 Preview tops the domestic charts. Wenxin 5.0 Preview 🔥 just snagged the #1 spot in China with a whopping 1451 points, according to the latest ranking data (AI News) from LMArena’s large model arena. It even outshone GPT-5.2 in creative writing! Boasting an astounding 2.4 trillion parameters and full multi-modality support, the official version is expected to drop in January.

2. DingTalk launches the world’s first AI Work OS. DingTalk just dropped the world’s first-ever AI Work OS (AI News) . Dubbed “Mulan,” this new version cleverly connects AI with the physical world. Its companion hardware, DingTalk Real, also made a simultaneous debut. CEO Chen Hang even joked that if things go sideways, you can just unplug the power directly (AI News) ! Now that’s what we call a hardcore understanding of AI safety. 💀

3. MiniMax unveils the strongest open-source coding model. MiniMax has just dropped its open-source coding M2.1 model (AI News) , specifically engineered for real-world coding scenarios. This bad boy absolutely crushed it in the VIBE-bench test, snagging an impressive 88.6% high score. It even outperformed closed-source models like Gemini3Pro! Seriously, this is currently the 💪 strongest open-source model for the agent era.
   
   

4. Zhipu AI officially unveils the GLM-4.7 flagship. Zhipu AI has officially launched its GLM-4.7 model (AI News) , marking a major upgrade for the GLM series. This powerhouse has achieved some serious game-changing progress 🌟 in programming and logical reasoning. It supports a massive 200K super-long context and even tool calling (AI News) . Plus, with its open-source weights, local deployment is now a go! 🏡
   
   

5. Claude Code gets native LSP protocol support. Claude Code now natively supports the LSP protocol (AI News) ! This means agents can now grab more precise code jump information. Users are already diving into a lively debate 🤔 about its practicality. While it’s seen as a big step 🚀 toward reducing context dependency, there’s still some room for improvement in terms of full functionality.

Frontier Research

1. Alibaba’s RTPurbo achieves 5x inference compression. Alibaba’s team just came up with RTPurbo ⚡, a new compression scheme (AI News) that only needs a mere 15% of full Attention for inference. This bad boy supercharges long-text inference speed by a whopping 5 times! It basically tackles the painful issue of pricey long-sequence computation head-on. Plus, the related code is already open-sourced on Huggingface. 📦
   
   

2. StoryMem enables coherent long video generation. StoryMem has unveiled a fresh paradigm for long video generation (AI News) . It leverages explicit visual memory 🧠 to keep shots super consistent, effectively transforming single-shot models into multi-shot storytellers. The result? Coherent minute-long video generation! And get this, the aesthetic quality scores are seriously off the charts. ✨

3. A new framework for detecting AI-generated images by fusing uncertainty. Check out this cool new framework for image (AI News) detection that focuses on AI-generated content! It ingeniously combines multiple uncertainty 🤔 metrics and uses a particle swarm optimization algorithm to pinpoint rejection thresholds. The kicker? It boasts a 70% interception rate 🛡️ for unseen generators and effectively fends off adversarial attacks. Pretty neat, huh?

Industry Outlook & Social Impact

1. ModelBest secures hundreds of millions in new funding round. ModelBest has just inked a funding agreement (AI News) worth hundreds of millions! The fresh capital will be pumped into R&D for efficient edge-side 📱 large models. Their MiniCPM model has already landed in the automotive sector 🚗, and they’ve already partnered up with giants like Huawei and Volkswagen. This is all about accelerating the popularization of edge AI! ✨

2. Supabase data exposure risks spark concern. Supabase data leak risks are currently sparking heated discussion (AI News) . Turns out, a ton of developers didn’t enable RLS security policies 🛡️, leaving user tables totally exposed in Public mode. To make matters worse, LLM code generation tools only amplified this configuration blunder. Even the clarity of the platform’s documentation has been catching some serious flak. 😬

Open Source TOP Projects

1. rendercv: A Typst-based resume generator. This is rendercv, a resume generator (AI News) based on Typst. It’s specifically crafted for academics and engineers 🧑‍💻. The best part? You can generate stunning PDFs without any fuss! It’s already racked up ⭐ 5.8k Stars.

2. LEANN: The RAG framework for everything. Meet LEANN, the framework that lets you run RAG for everything (AI News) right on your personal device! It enables lightning-fast and 100% 🔒 private applications, while also saving a massive 97% storage space. This project has already snatched ⭐ 5.3k Stars.

3. Yuxi-Know: A knowledge graph platform. Yuxi-Know is a knowledge graph platform (AI News) that teams up with LightRAG. It seamlessly integrates LangChain v1 and web retrieval, supporting tool calling and a variety of models. 🛠️ This project currently boasts ⭐ 2.7k Stars.

Social Media Shares

1. Spring AI boosts debugging capabilities. Spring AI’s feature iteration is seriously getting more complete (AI News) ! The new version introduces a Tool Parameter Enhancer ⚙️ mechanism, capable of capturing LLM’s internal inference metadata. This supercharges both debugging and observability big time! 🚀
   
   

2. Zhihu releases its annual AI product list. Zhihu’s annual AI product list 📊 has sparked heated discussion (AI News) . Doubao and Deepseek basically carved up the domestic Chinese market 🇨🇳. Gemini made a strong push at year-end, proving both powerful and low-cost. Meanwhile, Cursor officially kicked off a new era of Agent Programming! 🚀
   
   

3. Independent developer shares 2025 year-end reflections. An independent developer just shared some raw personal reflections (AI News) on 2025. Despite hitting over $100K in monthly income 💸, they still felt a bit lost. They delved into the whole “where does personal value go in the AI era?” question, refused to pretend to be free, and ultimately decided to close off options. Deep stuff! 🤯
   
   

4. Using Claude Skills to automate dating. Someone actually built a cyber matchmaker (AI News) using Claude Skills! It connects to their phone, scrolls through 200 videos of “beauties” daily 📱, automatically likes, comments, and even works to find partners. Rumor has it, its operating behavior is pretty much indistinguishable from a real human! Mind blown. 🤯

5. A useful Prompt for optimizing Claude Skill. Wanna optimize your Claude Skill? Someone shared a super useful Prompt (AI News) that lets the model check code against best practices 🔍. It automatically sniffs out vulnerabilities and even irons out the tiny details. The actual test results? Seriously significant optimization effect! ✨
   
   

AI Daily News Audio Version

🎙️ Xiaoyuzhou	📹 Douyin
Afterlife Tavern	Media Account