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 learn about innovative AI product applications.
Fresh AI products Click to learn more:https://app.aibase.com/zh
1. MiniMax released its first financial report after listing
MiniMax has released its first annual financial report after listing, showcasing significant progress in its AI platform strategy and financial performance. Despite facing risks such as copyright litigation, its global market expansion and cost advantages have brought tremendous growth potential.
【AiBase Summary:】
🚀 MiniMax's total revenue in 2025 reached $790.4 million, a 158.9% increase year-over-year.
📉 Annual loss of $1.872 billion, mainly due to revaluation of fair value of financial liabilities.
🌍 Overseas revenue accounted for over 70%, with Talkie having 236 million users in over 200 countries.
2. Alibaba Tongyi Qianwen releases open-source Qwen3.5 small model series: Edge devices can run multimodal agents
The Alibaba Tongyi Qianwen team launched the Qwen3.5 small model series, including four lightweight models of 0.8B, 2B, 4B, and 9B, supporting multimodal processing and suitable for edge devices and real-time interaction scenarios. The series covers a complete size gradient from 0.8B to 397B-A17B, further improving the open-source ecosystem.
【AiBase Summary:】
🚀 The Qwen3.5 small model series includes four lightweight models of 0.8B, 2B, 4B, and 9B, optimized for edge devices.
🧠 Supports multimodal capabilities, featuring improved model architecture and scalable reinforcement learning training.
🌐 Models are available on Hugging Face and ModelScope, developers can download and experience them immediately.
Details: https://huggingface.co/collections/Qwen/qwen35
3. Google NotebookLM officially launches infographic customization features
Google NotebookLM officially launched the infographic customization feature, offering 10 preset style options and supporting custom editing. Users can easily adjust visual presentation methods. This feature quickly converts complex source materials into high-quality infographics with a single click, greatly lowering the production barrier, suitable for learning summaries, knowledge sharing, and teaching demonstrations.
【AiBase Summary:】
🎨 10 preset style options, fully customizable editing support.
🔄 Convert complex source materials into infographics with one click.
💡 Enhances creativity and practicality, suitable for learning summaries, knowledge sharing, etc.
4. Claude Code official voice mode launched: Input /voice, press space to speak, AI programming efficiency soars
Claude Code launched the voice mode feature, allowing developers to quickly issue instructions through natural speech, improving programming collaboration efficiency. The feature has been gradually rolled out to users, featuring real-time voice transcription, hands-free efficient collaboration, and native integration with stable reliability.
【AiBase Summary:】
✅ Real-time voice transcription: Voice content is instantly converted into editable text, making it easy to correct or add details immediately.
✅ Hands-free efficient collaboration: Suitable for describing complex logic, reporting bugs, refactoring code, or planning projects, allowing developers to focus on thinking rather than typing.
✅ Native integration with stable reliability: As an official native feature of Claude Code, it seamlessly integrates with existing CLI/terminal workflows, offering better compatibility than third-party voice solutions.
5. GPT-5.4 unexpectedly “exploded”! GitHub source code leaked, OpenAI's top card: 2 million long context + stateful AI, ending the “goldfish memory” era?
The article reveals potential updates of the upcoming OpenAI GPT-5.4 model, including a 2 million context window, stateful AI, and pixel-level image processing capabilities, indicating a major breakthrough in AI technology.
【AiBase Summary:】
🧠 GPT-5.4 introduces stateful AI, enabling cross-session memory and workflow retention.
🖼️ Supports full-resolution raw byte reading, improving image recognition accuracy.
🤖 Emphasizes automatic agent execution, driving AI from chat assistants to digital employees.
6. One sentence becomes a hit short film! Zopia makes a big debut: Multi-Agent collaboration to deliver movie-quality footage in one click, 24-hour unmanned automated production
Zopia, the world's first end-to-end AI video director agent, achieves automated movie-level production through multi-agent collaboration, significantly improving the quality and efficiency of video content and bringing revolutionary changes to short film creation.
【AiBase Summary:】
🎬 Zopia realizes full-process automated production from script to final output through multi-intelligent agent collaboration.
🎥 Deeply optimized for Kling3 and Vidu Q3, enhancing realism and character consistency in real scenes.
🔄 Supports natural language interaction, allowing users to adjust plots or styles at any time, achieving "chat and edit simultaneously."
Details: https://zopia.ai/
7. QM releases 2025 AI Application Ranking: Doubao, DeepSeek, Yuanbao, Afu, Qianwen rank top five
The Quest Mobile report "Core Report on the Development of the 2025 AI Application Layer" shows that by December 2025, the top five monthly active users among all AI-native apps are Doubao, DeepSeek, Yuanbao, Ant Afu, and Alibaba Qianwen. Ant Group's Lingguang also entered the top ten. The report analyzes that in 2025, domestic AI applications have shifted from "general coverage" to "scenario penetration." Among the top ten AI apps, six are general AI and four are specialized AI for vertical scenarios. ByteDance and Alibaba AI applications occupy six of the top ten positions.
【AiBase Summary:】
🧠 The top five monthly active users of AI-native apps are Doubao, DeepSeek, Yuanbao, Ant Afu, and Alibaba Qianwen.
📊 In 2025, domestic AI applications have shifted from "general coverage" to "scenario penetration," with six general AI and four specialized AI for vertical scenarios among the top ten AI apps.
👥 ByteDance and Alibaba AI applications occupy six of the top ten positions, with three apps each listed.
8. DeepSeek V4 Lite quietly evolves: A 200-billion parameter small model performs impressively, rivaling top international models
DeepSeek V4 Lite, with a 200-billion parameter scale, demonstrates performance comparable to top closed-source models after continuous iterations, becoming a new benchmark for domestic large models. Its official release may have a significant impact on the global AI competition landscape.
【AiBase Summary:】
🚀 Small parameters, big power: Achieving performance comparable to top international closed-source models (such as Sonnet4.6) with only about 200 billion parameters.
📈 Quiet evolution: After multiple iterations in late February and early March, the model has significantly improved in programming, front-end development, and aesthetic capabilities.
🏁 Domestic new benchmark: It has reached the top level (SOTA) of domestic large models in multiple unofficial evaluations, triggering widespread anticipation for the full version of V4.
