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, helping you understand technology trends and innovative AI product applications.

New AI products Click to learn more:https://app.aibase.com/en

1. ByteDance launches Seed3D2.0: Dual SOTA in Geometry and Texture, API launched on Volcano Engine

ByteDance's Seed team has released the next-generation 3D generation large model Seed3D2.0, marking a key step towards "production-level usability" in AI 3D content generation. The model has achieved significant breakthroughs in geometric accuracy and material realism, with core indicators reaching industry-leading levels, effectively solving industry pain points such as edge softening, topological distortion, and lack of physical consistency in materials. Technically, Seed3D2.0 introduces a two-stage DiT architecture from coarse to fine, which greatly improves the ability to restore sharp edges and thin-walled structures by decoupling overall structure and geometric details. In texture generation, the model upgrades to a unified PBR generation architecture and combines MoE architecture with VLM prior knowledge to achieve precise decomposition of physical properties at high resolution. Evaluation data shows that Seed3D2.0 has human preference rates of over 80% in geometry generation and 69% in texture generation, significantly outperforming existing mainstream models. In addition, Seed3D2.0 further expands the application boundaries, supporting component-level decomposition, joint modeling, and complex scene combinations, allowing generated 3D content to seamlessly integrate with physics simulation engines like Isaac Sim. Currently, the technical report is publicly available, and the API is also launched on Volcano Engine. As 3D generation moves from single models to interactive, embodied intelligent scene-level construction, the release of Seed3D2.0 not only accelerates the digitalization process of industrial manufacturing and simulation training but also provides key infrastructure support for the large-scale implementation of embodied intelligence.

image.png

【AiBase Summary:】

📐 Seed3D2.0 achieves breakthroughs in geometric accuracy, reaching industry-leading levels.

🎨 Texture generation adopts a unified PBR architecture, enhancing the ability to accurately decompose physical properties.

⚙️ Supports component-level decomposition and complex scene combinations, seamlessly integrating with physics simulation engines.

Details: https://seed.bytedance.com/seed3d_2_0

2. Xiaomi MiMo-V2.5 Shakes the Market: 4.3-Hour Hand-Crafted Compiler, Long-Range Agent Achieves "Perfect" Milestone

Xiaomi officially announced the public beta test of its Xiaomi MiMo-V2.5 series large model, which includes MiMo-V2.5, V2.5-Pro, and accompanying TTS and ASR models, marking a comprehensive evolution of Xiaomi's large models from "usable" to "user-friendly." MiMo-V2.5-Pro has reached a level where it can compete with global top models such as Claude Opus4.6 and GPT-5.4 in general agent capabilities and software engineering dimensions, demonstrating extremely high instruction following and self-correction abilities. At the same time, MiMo-V2.5 is positioned as a native multi-modal agent, supporting integrated conversion of seeing, hearing, and reading, and has made significant optimizations in Token efficiency. In addition, Xiaomi also launched an optimized Token Plan to lower the usage threshold.

image.png

【AiBase Summary:】

🔥 MiMo-V2.5-Pro excels in complex tasks, completing a SysY compiler task in just 4.3 hours.

📊 MiMo-V2.5 approaches top closed-source models in video understanding and chart analysis, with Token efficiency improved by 42% to 50%.

💡 Xiaomi introduced an optimized Token Plan to reduce costs, including reduced rates and time-based discounts.

3. Alibaba Qwen Again "Turns the Table": Qwen3.6-27B Officially Open-Sourced, Programming Capabilities Show "Small vs. Big" Strategy

The Alibaba Cloud Qwen team released Qwen3.6-27B, which performs well in programming capabilities and multimodal processing, while deeply adapting to developer workflows, providing an accurate coding assistance environment for developers worldwide.

image.png

【AiBase Summary:】

🧠 Qwen3.6-27B performs excellently in programming benchmark tests, surpassing previous models.

🖼️ Supports mixed input of images, videos, and text, with strong multimodal processing capabilities.

🛠️ Deeply adapts to developer workflows, offering API calling services and local deployment support.

4. Lingguang Launches "Divide One Billion Incentive Plan", with Up to 10,000 Yuan per Creator

Lingguang App launched the "Divide One Billion Incentive Plan" to support high-quality flash apps and outstanding creators. The plan invests up to 1 million yuan daily to reward 10,000 high-quality flash apps, and up to 1 million yuan weekly to reward 100 outstanding creators. At the same time, Lingguang App also launched Lingguang Circle, connecting the entire process from generation to sharing.

image.png

【AiBase Summary:】

💰 Daily investment of up to 1 million yuan to reward 10,000 high-quality flash apps, each receiving 100 yuan.

🏆 Weekly investment of up to 1 million yuan to reward 100 outstanding creators, each receiving 10,000 yuan.

🌐 Lingguang App launched Lingguang Circle, connecting the complete process from generation to sharing.

5. Gaode Launches Automotive Travel AI Agent: Based on Qwen Model for Active Intention Understanding

Gaode Map launched an automotive travel AI Agent, marking a paradigm shift in car navigation from passive response to active intention understanding. The system is based on the Qwen model, building a dual-engine architecture of language brain and space brain, capable of parsing daily language and matching path resources, solving the pain point of "people adapting to systems" in smart cabins. Technically, its breakthrough lies in the proactive service capability of perception of time, space, and context, such as real-time monitoring of remaining battery power and automatically inserting charging nodes, or calculating detour plans before traffic accidents occur. This iteration indicates that the focus of smart cabin development will shift toward intention understanding and capability collaboration, and the relationship between car manufacturers and map suppliers will evolve from SDK integration to deep agent capability fusion. The emergence of AI Agent-native cabins means that the vehicle scenario is becoming a core killer application for large models. Gaode's move not only redefines the logic of human-machine interaction but also lays the infrastructure foundation for personalized travel services in the future of autonomous driving by integrating spatial data and semantic understanding.

image.png

【AiBase Summary:】

🚗 Gaode's automotive travel AI Agent is based on the Qwen model, achieving active intention understanding.

🧭 The system has reasoning and memory capabilities, capable of completing complex journeys with one sentence and dynamic route optimization.

🌐 AI Agent-native cabins have become a core application for large models, redefining human-machine interaction logic.

6. Tencent Launches Its Latest Open Source Language Model Hy3 Preview, Leading the Trend of the Intelligent Era

Tencent launched its latest open source language model Hy3 Preview, which has significant improvements in complex reasoning and instruction following, and has already been launched in multiple Tencent products. At the same time, Tencent Cloud also provides competitive API pricing and various package options to meet users' different needs.

image.png

【AiBase Summary:】

🧠 Hy3 Preview is Tencent's latest open source language model with 29.5 billion total parameters and 256K context support.

💡 The model has significant improvements in complex reasoning and instruction following capabilities and has already been launched in multiple Tencent products.

💰 Tencent Cloud launched competitive API pricing and various package options to meet users' different needs.

7. Dama Entertainment Joins Alibaba Tongyi to Launch "Miaoya" Platform: The First AI Fashion Toy Design Tool Based on Wan2.7

Dama Entertainment jointly launched the "Miaoya" platform with Alibaba Tongyi, based on Wan2.7 to create the first AI fashion toy design tool, marking a major breakthrough in AI technology in the fashion toy field.

image.png

【AiBase Summary:】

🎨 Miaoya platform integrates Wan2.7 large model capabilities, achieving a closed-loop transformation from concept to physical creation in fashion toy design.

🚀 Miaoya launched the "Fashion IP Creation Camp" program, providing incubation resources and traffic support for emerging creators.

🌐 The platform introduces Z-generation aesthetic perspectives through cooperation with universities, expanding the original boundaries of the fashion toy ecosystem.

Details: https://www.miaoyaai.cn/login

8. The Era of Large Model Ticket Purchasing Has Begun: China Eastern Airlines Is the First to Connect with Tongyi Qwen for Full-Process AI Services

The article introduces the collaboration between China Eastern Airlines and Alibaba Tongyi, marking the entry of AI ticketing services into a full-process era. Users can input their travel plans in natural language, and the AI will complete queries, comparisons, and ticket purchases. AI orders are directly supported by China Eastern Airlines for refunds, changes, and cancellations, avoiding third-party risks.

image.png

【AiBase Summary:】

✈️ China Eastern Airlines collaborated with Tongyi Qwen to realize full-process AI ticketing services.

🔄 Users can input their travel plans in natural language, and AI completes queries, comparisons, and ticket purchases.

🔒 AI orders are directly supported by China Eastern Airlines for refunds, changes, and cancellations, avoiding third-party risks.