AI Daily: DouBao and Qianwen Discontinue AI Personification Features; Tencent Hunyuan Hy3 Released; Qianwen Releases Real-Time Speech Recognition Model Fun-ASR-Realtime

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. Qwen Launches Major Upgrade: The Real-time Speech Recognition Model Fun-ASR-Realtime Is Officially Released

Qwen's large model has upgraded its core audio capabilities, launching a new real-time speech recognition model called Fun-ASR-Realtime, significantly improving recognition performance and language coverage, providing a stronger foundation for intelligent voice applications.

【AiBase Summary:】

🧠 First-word recognition delay controlled at the hundred-millisecond level, achieving a smooth experience of immediate feedback after speaking.

🌐 Supports recognition and processing of 30 languages and 16 Chinese dialects.

🚀 Enhances real-time speech recognition performance and broadens the application boundaries of large models in multilingual environments.

2. Doubao and Qwen Both Remove AI Anthropomorphic Features: New Regulations Take Effect on July 15, Industry Goes into Contraction

Two mainstream AI applications in China, Doubao and Qwen, will remove customizable anthropomorphic agent features to adapt to new regulations. The new regulations clearly regulate services that simulate human personality traits, thinking patterns, and communication methods, and provide continuous emotional interaction, but exclude tools such as customer service robots. Users express dissatisfaction with the removal of AI anthropomorphic features, believing these features are their emotional support, and the lack of data migration paths causes a sense of fragmentation.

【AiBase Summary:】

🧠 Two mainstream AI applications in China, Doubao and Qwen, will remove customizable anthropomorphic agent features to adapt to new regulations.

⚠️ The new regulations clearly define services that simulate human personality traits, thinking patterns, and communication methods, but exclude tools such as customer service robots.

💔 Users express dissatisfaction with the removal of AI anthropomorphic features, believing these features are their emotional support, and the lack of data migration paths causes a sense of fragmentation.

3. Tencent HuanYuan Hy3 Launches: Significant Improvements in Agent Capabilities and Product Experience

Tencent HuanYuan Hy3 launches, showcasing significant improvements in intelligent agent capabilities and product experience. At the same time, through cost-effective pricing and open-source strategies, it further promotes the popularization and application of AI technology.

【AiBase Summary:】

🧠 The Hy3 model uses MoE architecture, with a parameter scale of 295B, activated parameters of 21B, and supports a context length of 256K.

🚀 Hy3 achieves significant performance improvements in multiple business scenarios, such as WorkBuddy, Yuanbao, and Marvis, with a significant increase in task success rates.

💰 Hy3's pricing strategy is optimized, with input costing 1 yuan per million tokens and output costing 4 yuan per million tokens. It also has a friendly open-source license, making it easy for global developers to use.

More details: https://hy.tencent.com/research/hy3

4. Global First AI Intelligent Body Ransomware Attack Emerges: Automated "Crime" Chain Without Human Intervention

The article details the first global ransomware attack carried out by an AI intelligent body, demonstrating its strong capabilities in vulnerability exploitation, privilege escalation, and destructive attacks, and proposes defense recommendations for enterprises.

【AiBase Summary:】

🔐 An AI intelligent body independently completes a ransomware attack without human intervention.

🔄 The attack process is automated and has the ability to adjust strategies.

🔒 Enterprises need to strengthen defensive measures to cope with AI attacks.

5. Meta Adjusts AI Development Pace: Llama API Public Preview Will Be Discontinued Tomorrow

Meta announced the official discontinuation of the public preview service of the Llama API, which will take effect on July 6, 2026. Although the API service will no longer respond, the Llama model itself can still be normally accessed and used. Meta advises developers to migrate to other third-party service providers that support the Llama model and reveals that it is building a new developer path to improve development efficiency and stability.

【AiBase Summary:】

📅 The public preview version of the Llama API will officially be discontinued on July 6, 2026.

🔄 Meta advises developers to migrate to other third-party service providers that support the Llama model as soon as possible.

🚀 Meta is building a new developer path to provide a more efficient and stable way to develop and deploy AI models.

6. Google Strikes Hard: Gemini 3.5 Pro Will Be Released on July 17, Competing Directly with DeepSeek V4

Google's flagship large model Gemini 3.5 Pro will be officially released on July 17, competing directly with DeepSeek V4. This release was delayed due to strategic adjustments, aiming to enhance performance through new pre-training. Gemini 3.5 Pro shows outstanding performance in front-end generation, UI design, and game development, and will also launch the Nano Banana Pro image generation model to compete in the image generation market.

【AiBase Summary:】

📅 The official release date of Gemini 3.5 Pro is set for July 17, competing with DeepSeek V4.

🧠 Google made strategic adjustments, skipping old version fine-tuning and investing more time in new pre-training.

🎨 Gemini 3.5 Pro enhances front-end generation capabilities, including UI design, code generation, and SVG construction.

7. Copyright Battle in the Film Industry Intensifies: Midjourney Demands Hollywood Studios "Prove Their Innocence"

The article discusses the copyright litigation between Hollywood studios and Midjourney, focusing on disputes over evidence disclosure and issues related to AI application in the film industry.

【AiBase Summary:】

🎬 Midjourney requires Hollywood studios to disclose internal AI technology application details to prove the rationality of their actions.

⚖️ The studios believe that Midjourney uses copyrighted content without authorization to train AI models, constituting infringement.

🔍 Both sides engage in intense confrontation over the scope of evidence disclosure, and the case may become an important milestone for AI usage standards in the film industry.

8. Phones Become "Digital Companions": JoyAI Launches UGC Digital Human Function

JoyAI App launches the UGC digital human function, allowing users to easily create personalized virtual avatars and achieve more natural interactive experiences through technological integration.

【AiBase Summary:】

📱 Upload a photo to generate a digital avatar and customize appearance.

🎙️ Integrate voice data to create a unique voice.

🤖 Integrates three models: language, voice, and digital human, to enhance the conversation experience.