AI Daily: Meituan's Tabbit Browser Opens Beta Testing; Tongyi Releases Dual Voice Models; Notion Introduces Its First Open-Source Weight Model MiniMax M2.5

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 technical trends and learn about innovative AI product applications.

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

1. Light Years Away Tabbit Browser, a subsidiary of Meituan, opens public testing: integrated multi-model AI Agent, supports full automation of web operations

The release of Tabbit browser marks the transformation of browsers from traffic entry points to personal office brains with execution capabilities, providing a new direction for the popularization of AI technology and the reconstruction of productivity.

【AiBase Summary:】

🧠 Integrated multi-model AI Agent, achieving automated web operations.

🔄 Supports cross-system data transfer and automatic form filling.

💡 Introduces "Skill" and "Script" functions to enhance user efficiency.

2. Instantly "tune" voice with one sentence! Alibaba Tongyi releases a dual-model voice: supports FreeStyle natural language control

Alibaba Tongyi Lab launched two revolutionary voice generation models that support natural language instructions to control voice and scene, significantly improving the efficiency and quality of voice creation.

【AiBase Summary:】

🎙️ Fun-CosyVoice3.5 supports multilingual voice generation and fine-grained control, improving the accuracy and naturalness of voice expression.

🎧 Fun-AudioGen-VD can generate "person + scene" integrated audio based on descriptions, enhancing environmental immersion.

💡 Both models support "FreeStyle" instructions, allowing users to precisely control voice style or build complex audio scenes through natural language.

3. Breaking the closed-source monopoly! Notion introduces the first open-source weight model MiniMax M2.5: partnering with "national light" to reshape AI workflow, emphasizing "extreme cost-effectiveness"

Notion introduced the first open-source weight model MiniMax M2.5, breaking the long-standing monopoly of closed-source models and providing users with more cost-effective options. MiniMax M2.5 performs well in performance and inference speed, and has been deeply optimized for real office scenarios, highly matching Notion's "all-in-one workspace" characteristics. This move reflects Notion's core strategy for 2026 - model agnosticism, allowing users to freely switch models according to task requirements, further solidifying its position as the "digital brain" for individuals and enterprises in the AGI era.

【AiBase Summary:】

🧠 MiniMax M2.5, the first open-source weight model introduced by Notion, provides users with a more cost-effective option.

🚀 MiniMax M2.5 performs excellently in performance and inference speed, even surpassing some closed-source flagship models.

🔄 Notion introducing open-source models reflects its "model agnosticism" strategy, allowing users to freely switch models according to task requirements.

4. Creative one-stop integration! Google upgrades Flow AI tool: video, painting, and special effects fully integrated

Google has made large-scale functional expansions to the Flow AI tool, integrating image generation technology Whisk and ImageFX, providing creators with a unified AI creative studio.

【AiBase Summary:】

✨ Unified workflow: Users can complete the entire process from image generation to animation production in one interface.

🖼️ Driven by ImageFX: Utilize Google's latest DALL-E-level image model to generate high-quality materials.

🎨 Dynamic editing: Enhanced animation production features, supporting more refined action flow and special effects.

5. Alibaba's large model brand officially renamed "Qwen", achieving comprehensive integration of the brand matrix

Alibaba announced that it will unify its core AI business brand under "Qwen", marking a further focus on its AI strategy. Qwen large model performs strongly in both technical and market aspects, with its open-source models being the most active in the global developer community. At the same time, the user activity of the Qwen APP has significantly increased, showing the momentum of evolving into a national-level AI assistant.

【AiBase Summary:】

🧠 Qwen large model has excellent technical performance, with multiple models occupying the top four positions on the Hugging Face global open-source large model list.

📱 Qwen APP has high user activity, with daily active users reaching 73.52 million during the Spring Festival holiday, an increase of 940%.

🌐 Alibaba strengthens its leadership in the open-source ecosystem and consumer market through brand integration, promoting domestic large models to reach global top levels.

6. Samsung announces moving towards "AI factory" by 2030, Agentic AI as the core brain

Samsung announced that it will upgrade its global production system to an "AI factory" by 2030, driven by Agentic AI, realizing full-chain autonomy from raw materials to finished products.

【AiBase Summary:】

🧠 The AI factory will achieve full-process autonomy, improving production efficiency and consistency.

🔄 Digital twin technology will be used for real-time synchronization between virtual factories and physical production lines.

🤖 High-risk tasks will be completed by environmental safety robots combined with digital twin technology.

7. OpenAI suspected of acquiring GPT.com top domain name, now redirected to ChatGPT official website

OpenAI is suspected of completing the acquisition of the top domain name GPT.com, further consolidating its brand position in the generative AI field. The domain name now redirects to the ChatGPT official website, similar to the handling method after the previous acquisition of Chat.com. This move not only demonstrates OpenAI's defensive protection of brand assets but also shows its long-term strategic intent to build a full ecological entry point and deepen user mind share.

【AiBase Summary:】

✅ OpenAI is suspected of acquiring the GPT.com domain, strengthening its brand moat.

🔄 GPT.com has now redirected to the ChatGPT official website, consistent with the previous handling of Chat.com.

🌐 OpenAI builds a full ecological entry point through strategic domain name acquisitions, deepening user mind share.

8. Solving the long video retrieval problem! Peking University partners with OceanBase to create LoVR benchmark: selected for WWW 2026, opening up a new paradigm for "full video + segment-level" intelligent retrieval

The launch of the LoVR benchmark marks a new stage in multi-modal retrieval evaluation for long videos, solving the problem of complex semantic associations in long videos that traditional short video retrieval cannot handle, providing a reliable foundation for future video search engines and AI editing assistants.

【AiBase Summary:】

🎥 Full granularity coverage: supports full video and segment-level retrieval, meeting diverse needs.

📊 Large-scale high-quality annotation: data generated automatically by large models, achieving low-cost scalability.

🧠 Realistic scenario modeling: depicts challenges in long video retrieval, improving evaluation accuracy.