AI Daily: Baidu Launches Wenxin 5.0; Keling 2.5 Turbo Model Launches First and Last Frame Function; Weibo Launches VibeThinker-1.5B

Welcome to the "AI Daily" section! Here 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 innovative AI product applications.

New AI products click to learn more:https://app.aibase.com/zh

1. Control in Video Generation Gets a Major Upgrade! The Keling 2.5 Turbo Model Introduces the "First and Last Frame" Feature

The release of the Keling 2.5 Turbo model significantly enhances the controllability, stability, and consistency of AI video generation, offering a higher-quality solution for professional creative content production. It has made notable improvements in dynamic effects, text response accuracy, style preservation capabilities, and overall aesthetic effects, and introduced a new first and last frame feature that allows creators to more precisely control the start and end states of videos.

AiBase Summary:

🚀 The Keling 2.5 Turbo model significantly improves the controllability and stability of AI video generation.

🎨 Breakthroughs in dynamic effects, text response accuracy, and style preservation capabilities.

📽️ New first and last frame feature helps creators precisely control the start and end states of videos.

2. Baidu Launches the New Native Multimodal Large Model Wenxin 5.0

Baidu officially launched its latest native multimodal large model, Wenxin 5.0, at the Baidu World Conference on November 13, 2025. With 2.4 trillion parameters, it uses native multimodal unified modeling technology to simultaneously understand and generate text, images, audio, and video, demonstrating strong multimodal capabilities. Wenxin 5.0 performs well in multiple fields, with language and multimodal understanding capabilities comparable to top global models, and image and video generation capabilities leading globally. Users can experience the features through the Wenxin App, while developers and enterprise users can call API services through the Baidu Qianfan platform.

AiBase Summary:

🚀 Wenxin 5.0 uses native multimodal technology to support the understanding and generation of text, images, audio, and video.

📈 In authoritative benchmark tests, Wenxin 5.0's multimodal understanding capabilities are comparable to global top models, showing its great potential.

🌐 Users can experience the new model through the Wenxin App, and developers can use API services to promote intelligent application development.

3. Weibo Launches VibeThinker-1.5B, A Low-Cost AI Model Challenging Large Language Models

Weibo launched VibeThinker-1.5B, a large language model (LLM) with 1.5 billion parameters, which was finely tuned based on Alibaba's Qwen2.5-Math-1.5B and is freely available on Hugging Face, GitHub, and ModelScope. Despite its smaller size, it performs exceptionally well in math and code tasks, even surpassing the R1 model of DeepSeek with 67.1 billion parameters. Its post-training cost is only $7,800, far lower than the tens of thousands of dollars for similar models. VibeThinker-1.5B uses a training framework called "Spectrum-Signal Principle," allowing small models to achieve efficient reasoning capabilities.

AiBase Summary:

🧠 VibeThinker-1.5B is an open-source AI model from Weibo with 1.5 billion parameters, performing well and even surpassing large models.

💰 The post-training cost of this model is only $7,800, much lower than the tens of thousands of dollars for similar models.

🔍 Using the "Spectrum-Signal Principle" training framework, small models can achieve efficient reasoning, enhancing the competitiveness of small models.

Details: https://huggingface.co/WeiboAI/VibeThinker-1.5B

4. OpenAI Unveils GPT-5.1: Faster, More Accurate, and More Human-Like Personalized AI Assistant

OpenAI launched GPT-5.1 to enhance the flexibility, response speed, and personalized experience of ChatGPT. The new model shows significant improvements in language expression, dialogue style adaptability, and emotional perception, while introducing adaptive reasoning functionality to meet different task needs.

AiBase Summary:

🚀 GPT-5.1 improves response speed and language clarity, making conversations more natural.

🧠 New adaptive reasoning function adjusts processing time according to the complexity of the question.

🎨 Offers multiple communication styles for enhanced personalization.

5. Fei-Fei Li’s World Labs Releases First Commercial 3D World Model Marble, Supporting Multiple Input Generations

Fei-Fei Li’s World Labs released the first commercial 3D world model, Marble, which supports generating editable 3D environments using various input methods and includes AI editing features. It is compatible with mainstream VR devices and is applicable in areas such as game development and film special effects.

AiBase Summary:

🌟 Marble is the first commercial 3D world model, supporting multiple inputs to generate editable environments.

🎮 The product includes AI editing tools, allowing users to design and customize 3D scenes more conveniently.

🕶️ Marble is compatible with mainstream VR devices, enabling users to immediately experience generated 3D worlds.

Details: https://marble.worldlabs.ai/

6. Northeastern University Opens a "Nuclear Bomb" for Multilingual Translation! NiuTrans.LMT Supports 60 Languages and 234 Directions, Major Breakthrough in Low-Resource Language Translation

The NiuTrans.LMT large model developed by Northeastern University has made a major breakthrough in multilingual translation, supporting 60 languages and 234 translation directions, especially achieving significant progress in low-resource languages. Its dual-center architecture avoids secondary distortion, improving cross-cultural interaction efficiency and accuracy.

AiBase Summary:

🧠 Dual-center architecture breaks the English monopoly, supporting bilingual core translation.

🌐 Three-layer language coverage ensures efficiency and fairness, enhancing low-resource language translation capabilities.

🚀 Two-stage training tops FLORES-200, with excellent performance.

Details: https://github.com/NiuTrans/LMT

7. Google Gemini Live Voice Upgrade! Adjust Speed at Will, Choose Accent Freely, ChatGPT Voice Mode Faces Strongest Challenge

The upgrade of Google Gemini Live voice functions pushes AI conversation to a new level through five core capabilities, providing users with a more natural and personalized interactive experience.

AiBase Summary:

🗣️ Speed changes in real-time with voice commands, supporting personalized language training.

😊 Emotion perception, adaptive tone, improving the conversation experience.

🎭 Customizable accents, making conversations more interesting.

8. Alibaba's "Qianwen" Project Secretly Launched: Based on Qwen Model, Full Competition Against ChatGPT in Consumer AI Future

Alibaba has launched a secret project called "Qianwen," aiming to create a personal AI assistant under the same name, fully competing against ChatGPT. This marks Alibaba's official entry into the global top-tier competition in AI applications and elevates C-end AI applications to a strategic core.

AiBase Summary:

🚀 Alibaba launched the "Qianwen" project, creating a personal AI assistant to fully compete against ChatGPT.

💡 Leveraging the excellent performance and international influence of the Qwen model, Alibaba aims to win the AI competition.

📈 Alibaba is pushing C-end AI applications to the strategic core, focusing on the consumer market.