Welcome to the "AI Daily" section! This is your guide to exploring the world of artificial intelligence every day. Every day, we present you with the latest content in the AI field, focusing on developers to help you understand technology trends and innovative AI product applications.

Fresh AI products Click to learn more:https://app.aibase.com/zh

The MiMo large model charging function has been launched, and the paid era is about to begin!

Xiaomi officially announced that the open-source large model MiMo V2Flash API has launched a charging function and will soon enter the paid mode; at the same time, it gives all users free quotas, and can still be used for free before the billing system starts. Domestic pricing: input 0.7 yuan per million tokens, cache hit 0.07 yuan per million tokens, output 2.1 yuan per million tokens; overseas pricing: input 0.1 USD per million tokens, cache hit 0.01 USD per million tokens, output 0.3 USD per million tokens.

image.png

AiBase Highlights:

💰 Charging function is available, and the billing system will start soon, but it is still possible to use for free now.  

🌍 Differentiated pricing at home and abroad, multiple payment channels reduce barriers.  

⚡ 309 billion parameter open source giant, leading in speed and performance.

Details link: https://www.chinaz.com/ainews/24797.shtml

Meitu Show秀 tops the AI image export benchmark, AI Flash Light ignites global creation trend, 2025 AI image enters the era of high-quality products

In 2025, Meitu Show秀 was continuously selected as one of the top 100 AI flagship products with core functions such as "AI Flash Light", ranked in the top ten of the Qixmai Data "Powerful AI Product App List", won the title of "Annual Powerful Export Application", and became a model of domestic AI globalization. "AI Flash Light" allows mobile phone low-light photos to present a professional texture through intelligent light reconstruction and detail enhancement, triggering a download and usage boom in emerging markets such as Southeast Asia, Latin America, and the Middle East.

AiBase Highlights:

🏆 Top export list, AI Flash Light becomes a global hit.  

🌄 From "one-click beautification" to high-quality creation, AI images enter a refined era.  

🌐 Template + cross-cultural adaptation, building a global digital creative platform.

Details link: https://www.chinaz.com/ainews/24800.shtml

MiniMax launches "Expert Agent" desktop version, building an AI-native workspace, users can customize professional-level intelligent assistants

MiniMax released the desktop version of "MiniMax Agent Desktop App" and the web version 2.0, introducing the concepts of "Expert Agent" and "AI-native workspace" for the first time. Users can create advanced AI assistants with domain knowledge and skills, and deeply customize the operation logic locally, achieving "one person, one intelligent agent".

AiBase Highlights:

🖥️ Desktop + cloud workspace, local customization of personalized AI environment.  

🧑‍💼 One-click generation of "Expert Agent" using natural language, covering scenarios such as law, finance, and code.  

🚀 Free trial for a limited time, moving towards the entry point of "personal AI operating system".

Details link: https://www.aibase.com/zh/news/24803

Kimi large model is about to be upgraded! Open-source new star surpassing GPT-5 is coming

Yue Zhi Anmian announced the launch of a new trillion-parameter Kimi large model, which has already started internal testing for enterprise certified users. The industry guesses that the new model is K2.1 or K2.5, and will continue to use Kimi Delta Attention linear attention mechanism, comprehensively upgrading short and long text processing, aesthetics, and values. Last year, K2 and K2Thinking have achieved SOTA results in multiple benchmark tests and surpassed GPT-5, while the API pricing is lower than that of GPT-5.

Kimi AI, Yue Zhi Anmian

AiBase Highlights:

🚀 New Kimi with trillion parameters enters internal testing, performance aims for GPT-5+.  

🔍 Linear attention + value design, breaking away from homogenized competition.  

💰 Valuation 4.8 billion USD, hundreds of millions of cash investment in AI programming and multimodal.

Details link: https://www.aibase.com/zh/news/24791

The video AI has reached the "tipping point": Keling AI dominates 40 countries, revenue reaches 140 million USD in 2025

Keling AI, under Kuaishou, had monthly active users exceeding 12 million in early 2026, with a 350% increase in app-side paying users. The O1 multimodal video model and the 2.6 version "audio-video co-generation" capabilities released at the end of 2025, combined with the "action control" feature launched in January, triggered a global "pet dancing" virus spread. In 2025, the total revenue reached 140 million USD, doubling the target of 60 million USD; in December alone, revenue exceeded 20 million USD, and daily revenue increased by 30% month-over-month in January.

AiBase Highlights:

🎬 Monthly active users 12 million, paying users up 350%.  

💵 2025 revenue 140 million USD, over 133% of expectations.  

🌍 40 countries' download leaderboard, Kuaishou AI is revalued by Wall Street.

Details link: https://www.aibase.com/zh/news/24813

ChatGPT becomes "more mature": OpenAI launches age detection system, strictly controlling minors from accessing inappropriate content

OpenAI has introduced the "age prediction" feature on the consumer version of ChatGPT, automatically identifying users under 18 years old by combining account duration, activity periods, interaction patterns, and other behavioral signals. If the identification cannot be confirmed, a stricter security mode is enabled by default. Accounts identified as minors will automatically block five types of high-risk content: violence, bloodshed, dangerous challenges, sexual or violent role-play, self-harm, and unhealthy aesthetics. Adults mistakenly identified can quickly restore their permissions through third-party selfies, and parents can also set "quiet hours" and psychological distress reminders to protect teenagers' health comprehensively.

AiBase Highlights:

🛡️ Behavior signal AI predicts age, no longer relying only on registration information.  

🚫 Five categories of content are strictly blocked, minor's firewall is upgraded.  

👪 Parental controls + selfie verification, balancing safety and experience.

Details link: https://www.aibase.com/zh/news/24786

DeepSeek secret code exposed: "MODEL1" new architecture aims for February, programming capabilities evolve again

On the anniversary of DeepSeek-R1's release, GitHub FlashMLA code library was discovered by developers to have 28 "MODEL1" mysterious marks, including KV cache layout, sparsity processing, FP8 decoding, etc., showing that it is building a new architecture different from the current V3.2. Combined with previous published optimized residual connections (mHC) and "AI memory module (Engram)" papers, the industry guesses that this new flagship (or named DeepSeek V4) will be launched as early as mid-February, bringing stronger code generation and reasoning performance.

AiBase Highlights:

🔍 "MODEL1" code revealed for the first time, architectural-level restructuring.  

🧠 Residual connection + memory module new technology may be fully integrated.  

⏳ Countdown to the February new flagship, programming capabilities evolve again.

Details link: https://www.aibase.com/zh/news/24792

Medeo AI new version officially launched overseas: one-click script changes with natural language, everyone can shoot a movie

Medeo AI launched a new video proxy version overseas, centered around an intelligent proxy architecture, supporting dynamic editing of videos with natural language: adding or removing content and rewriting scripts do not require starting from scratch, like a "digital director" who splits scenes and locks rhythm in advance. The system automatically matches visuals, voiceover, and BGM, avoiding the randomness and inconsistency of traditional AI video tools.

QQ20260121-152317.jpg

AiBase Highlights:

🎞️ Natural language real-time script changes, zero-barrier iteration of videos.  

🎬 Pre-planned scenes + element auto alignment, significantly improving cinematic consistency.  

🌍 Overseas launch, everyone can easily shoot a movie with one click.

Details link: https://www.aibase.com/zh/news/24812