AI Daily: MiniMax Launches Music3 Music Model; Baidu GenFlow Announces Chinese Name Kuku AI; Zhipu Launches GLM-5.3

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.

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

1. MiniMax launches Music3 music model: Generate a 5-minute song with lyrics and descriptions

MiniMax-Music3 is a music generation model that can generate a complete 5-minute song based on lyrics and music descriptions. It uses a hierarchical autoregressive architecture, where a global language model and a local language model work together, responsible for long-range semantics and structural changes of the song, as well as fine-grained acoustic information completion. The model can maintain the musical theme, rhythm, vocal identity, and arrangement progress in long audio, and supports various music structures such as prelude, verse, and chorus.

AiBase Summary:

🎵 The global language model is responsible for the long-range semantics and structural changes of the song.

🔊 The local language model is responsible for predicting the acoustic codebook of each frame and supplementing fine-grained acoustic information.

🎼 The model can generate a 5-minute song containing multiple music structures.

More details: https://huggingface.co/MiniMaxAI/MiniMax-Music3

2. Baidu GenFlow officially announced its Chinese name "Kuku AI": Launch PC, web, mini-program, and enterprise versions

Baidu Wenshu Cloud Disk General Intelligent Agent GenFlow has officially announced its Chinese name "Kuku AI" and launched an independent product matrix covering various office scenarios. Data shows that GenFlow's monthly active users have exceeded 100 million, and its AI office monthly active users (MAU) exceed 25 million. This upgrade marks further strengthening of Baidu's layout in the commercialization of general intelligent agents, helping to expand its usage scenarios and service boundaries.

AiBase Summary:

📅 Baidu Wenshu Cloud Disk General Intelligent Agent GenFlow officially announced its Chinese name "Kuku AI"

📊 GenFlow's monthly active users have exceeded 100 million, and AI office monthly active users exceed 25 million

💻 Kuku AI has launched an independent product matrix, covering personal office, web use, mobile applications, and enterprise-level scenarios

3. Zhipu releases GLM-5.3: No base model change, only post-training, the throne of open-source programming is taken over

Zhipu released GLM-5.3, which significantly improves model performance through scaling in the post-training phase, enhances programming capabilities, and performs excellently in terms of security.

AiBase Summary:

🧠 The improvement of GLM-5.3 mainly comes from scaling in the post-training phase, not changing the base model.

💻 The new model significantly improves programming capabilities and becomes one of the strongest open-source models currently available.

🛡️ GLM-5.3 performs well in terms of security, showing potential for network defense scenarios.

4. OpenAI launches a new "Computer History" feature, allowing users to remember operations without screenshots

OpenAI has launched a new feature called "Computer History," aiming to enhance ChatGPT's contextual understanding and retrieval capabilities by tracking user interaction events, while ensuring user privacy and autonomy.

AiBase Summary:

🧠 The "Computer History" feature generates memory by tracking user interaction events, enhancing ChatGPT's contextual understanding.

🔒 This feature is disabled by default, and users can pause or exclude specific applications at any time, ensuring privacy and autonomy.

🌐 It has been rolled out to all applicable subscription users globally, and users in the European Economic Area, the UK, and Switzerland will gain access in the coming weeks.

5. Google launches Gemini 3.7 Flash: Coding performance fully upgraded, initial pricing reduced by 50%

Google launched Gemini 3.7 Flash, achieving significant breakthroughs in coding performance and AI Agent scenario functions, with comprehensive performance surpassing several competing models. At the same time, the initial pricing of the model was greatly reduced, promoting the accelerated commercial application of high-value large models.

AiBase Summary:

🧠 Gemini 3.7 Flash has made significant upgrades in coding performance, with a FrontiersCode benchmark score of 43.6%

📊 It performs well in web development, document understanding, and business process automation, surpassing Claude Sonnet 5 and GPT-5.6 Terra in comprehensive performance.

💰 Initial pricing reduced by 50%, $0.75 per million input tokens, $3.75 per million output tokens

6. App Store now has AI fraud apps: Fake screenshots get 4.9 stars, developers accuse Apple's review process is ineffective

The article reveals that there are fraudulent apps in the App Store using AI-generated fake screenshots and ratings, pointing out that Apple's review mechanism has loopholes, and emphasizing that such apps may threaten user privacy.

AiBase Summary:

🍎 A Safari extension named TabControl faked screenshots to reach 12th position on the Mac App Store paid list.

💣 The app's fake rating and AI-generated promotional images are clearly unreasonable, leading developers to question Apple's review process.

🔒 Safari extension permissions may threaten user privacy, and such fraud apps pose serious risks once installed.

7. DeepSeek V4 Pro lands on Silicon Flow: 1 million context window with three levels of inference intensity, cache hit cost only $0.44 per million tokens

DeepSeek-V4-Pro-0813 is officially launched on the Silicon Flow platform. Its core selling points include a 1M context window, three levels of inference intensity, and emphasis on coding, tool calls, and agent workflows. The pricing structure is clear, with cache hit costs as low as $0.44 per million tokens, providing a more specific cost anchor for long-text scenarios.

AiBase Summary:

🧠 1M context window improves processing power

⚙️ Three levels of inference intensity meet different needs

💰 Cache hit cost is as low as $0.44 per million tokens

8. JD Qixian Coffee launches the world's first 24-hour unmanned store: robot-made, AI recommendation, located in Beijing Galaxy Soho

JD Group's Qixian Coffee announced the opening of the first 24-hour smart unmanned coffee shop in Beijing Galaxy Soho, featuring a full-day, fully automated concept, with robots and AI taking over the entire production process and offering personalized customization services. At the same time, the AI recommendation function will intelligently recommend caffeine intake based on the user's physical condition, reflecting the further development of intelligent retail scenarios.

AiBase Summary:

☕ Opening of the first 24-hour intelligent unmanned coffee shop, robot-made coffee

🤖 AI personalized recommendation function launched, recommending caffeine intake based on physical condition

📈 Qixian Coffee is trying the unmanned + AI model, exploring new directions for intelligent retail scenarios