Welcome to the 【AI Daily】column! This is your daily guide to exploring the world of artificial intelligence. Every day, we present you with the hottest AI news, focusing on developers and helping you understand technology trends and innovative AI product applications.
New AI Products Learn More: https://top.aibase.com/
1. The Priciest Yet! OpenAI Launches Upgraded AI Model o1-pro, Generating Price is Ten Times That of o1
OpenAI recently launched its next-generation AI model, o1-pro, aiming to provide superior reasoning capabilities. However, its high price has drawn considerable attention. The input and generation costs of o1-pro are double and ten times that of GPT-4.5 and the standard o1, respectively. Despite this, OpenAI has high hopes for its performance, believing it can meet developers' needs for complex tasks.
【AiBase Summary:】
💡 OpenAI launches new AI model o1-pro, aiming to enhance reasoning capabilities.
💰 o1-pro is extremely expensive; input costs are double that of GPT-4.5, and generation costs are ten times that of the standard o1.
🤔 Early user feedback on o1-pro is mixed, but it shows more reliable performance on coding and mathematical problems.
2. Motion Controllable! Step-Video-TI2V, a Text-to-Video Model from Jieyue Xingchen, is Open-Sourced
Shanghai Jieyue Xingchen Intelligent Technology Co., Ltd. has launched the Step-Video-TI2V model, a significant innovation in the text-to-video field. Based on the 30B-parameter Step-Video-T2V, this model can generate high-quality videos with controllable motion amplitude and camera movement, making it particularly suitable for animation creation and short video production. By optimizing consistency and dynamism, the model offers creators more flexibility to meet diverse size and effect requirements.
【AiBase Summary:】
🚀 Step-Video-TI2V model, based on 30B parameters, generates 5-second, 540P resolution videos with controllable motion amplitude and camera movement.
🎨 The model excels in anime effects and is suitable for animation creation and short video production, supporting multiple size generations.
🔧 By introducing image conditions and the AdaLN module, the consistency and dynamic control of generated videos with the original image are improved.
Details link: https://yuewen.cn/videos
3. Tencent HunYuan Makes New Moves! New Inference Model T1 to be Released on the Evening of March 21
Tencent HunYuan announced that its new inference model T1 will be officially released on March 21st, marking a technological iteration and product upgrade in the field of large AI models. Simultaneously, the Tencent HunYuan large model has entered the Chatbot Arena global Top 15 ranking for the first time, demonstrating its internationally leading technical capabilities. The outside world anticipates improvements in the T1 model's reasoning capabilities, further solidifying Tencent's position in the global large model competition.
【AiBase Summary:】
🚀 Tencent HunYuan will release its new inference model T1 on March 21st, marking a technological upgrade.
🏆 Tencent HunYuan's large model has entered the Chatbot Arena global Top 15 ranking for the first time, showcasing its technical strength.
🌍 The outside world expects improvements in T1's reasoning capabilities, solidifying Tencent's position in global competition.
4. Cost Only One-Tenth! Open-Sora 2.0 Open-Source Video AI Achieves Commercial-Grade Quality
HPC-AI Tech recently launched Open-Sora 2.0, a revolutionary video AI system. Its training cost is only one-tenth that of traditional systems, and its output quality rivals commercial products. The system achieves significant training speed improvements through a three-stage training process and efficient autoencoders, although there are some limitations in resolution and video length. The launch of Open-Sora 2.0 may have a profound impact on the cost structure of the video AI field, driving competition between open-source and commercial systems.
【AiBase Summary:】
💡 Open-Sora 2.0's training cost is only $200,000, significantly lower than the millions of dollars for existing high-quality video generation systems.
⚙️ The system uses a three-stage training process and a video DC-AE autoencoder, providing 5.2 times faster training speed and over ten times faster video generation speed.
📈 Open-Sora 2.0's VBench score is only 0.69% lower than OpenAI's Sora, performing excellently in visual quality and prompt accuracy.
5. Boston Dynamics' Atlas Robot Makes Another Breakthrough: Movement Capabilities Approach Human Levels
Boston Dynamics recently showcased the latest movement capabilities of its humanoid robot, Atlas. Combining reinforcement learning and motion capture technology, Atlas can self-learn and exhibit more natural and flexible human-like movements. This technological breakthrough is considered to bring humanoid robots closer to real-world applications, especially in potential applications in industrial, medical, and rescue fields.
【AiBase Summary:】
🤖 Atlas achieves more natural human-like movements through reinforcement learning and motion capture technology.
🚀 This technological breakthrough improves the robot's adaptability and coordination in complex environments.
🌐 Boston Dynamics' collaboration with the RAI Institute adds more possibilities for the commercialization of humanoid robot technology.
6. Explosive! Humanoid Robot Shows "Human Ceiling" Movements, Unitree G1 Completes the First Side Somersault, and Dares to Challenge Humans!
Unitree Robotics' G1 humanoid robot successfully completed a highly difficult side somersault and landed steadily, marking a major breakthrough in its robotic movement capabilities. This achievement not only demonstrates the G1's high reliability and success rate but has also attracted widespread attention from global technology enthusiasts. To further verify its capabilities, Unitree Robotics has launched a "Robot Side Somersault Human Challenge," encouraging humans to attempt this difficult movement, with the winner receiving a G1 robot or an equivalent prize.
【AiBase Summary:】
🤸♂️ Unitree Robotics' G1 robot successfully completes a side somersault, becoming the world's first humanoid robot to achieve this.
🏆 Unitree Robotics launches the "Robot Side Somersault Human Challenge," encouraging humans to attempt this difficult movement.
🌍 The competition has attracted the attention of global technology enthusiasts, anticipating the first person to successfully replicate the robot's side somersault.
7. Adobe Launches "Project Slide Wow," Turning Data into Eye-Catching PPTs with One Click
At Adobe's annual digital innovation conference, the "Project Slide Wow" project garnered significant market attention. This generative AI-powered tool aims to quickly transform raw customer data into engaging PowerPoint presentations, greatly simplifying the work of data analysts and marketers. By automatically generating high-quality slides and incorporating a built-in smart assistant, users can update and adjust presentation content in real time, ensuring accuracy and timeliness.
【AiBase Summary:】
✨ Generative AI tool quickly transforms raw data into high-quality PPTs, greatly simplifying the creation process.
🤖 A built-in smart assistant responds to user needs in real time, providing additional visualization and dynamic slide generation.
📊 Features real-time data updates to ensure presentation information is always current, improving enterprise decision-making efficiency.
8. Orpheus TTS: A New Generation TTS Model with Human-Like Emotional Expression
Orpheus TTS is a newly launched open-source text-to-speech model that has attracted widespread attention for its ultra-low latency and high emotional expression capabilities. The model performs exceptionally well in real-time conversation scenarios, providing natural and fluent voice output and greatly enhancing the experience of intelligent voice interaction. Its open-source nature also provides developers with more customization possibilities, and it is expected to become a benchmark in multiple fields in the future.
【AiBase Summary:】
⚡ **Ultra-low latency**: Default latency is approximately 200 milliseconds, compressible to 25-50 milliseconds through optimization, meeting real-time conversation needs.
🎭 **Emotional expression**: Voice output is natural and fluent, supporting a rich variety of intonation changes, enhancing the interaction experience.
🎙️ **Real-time output stream**: Supports streaming audio generation, ensuring synchronization of voice generation with input, suitable for various scenarios.
Details link: https://github.com/canopyai/Orpheus-TTS
9. LG Open-Sources EXAONE Deep Model, Claimed as South Korea's First Self-Developed Inference AI Model
LG AI Research recently open-sourced the EXAONE Deep inference AI model, marking the entry of AI into a new era of proactive AI. With 32 billion parameters, this model demonstrates exceptional reasoning capabilities, particularly excelling in logical reasoning and mathematics, achieving a score of 94.5 in the Korean College Entrance Exam mathematics section—comparable to a top student.
【AiBase Summary:】
🧠 EXAONE Deep is South Korea's first self-developed inference AI model, capable of independently formulating hypotheses and verifying inferences.
📊 The 32-billion parameter EXAONE Deep excels in logical reasoning and mathematics, particularly achieving 94.5 points in the Korean College Entrance Exam.
📱 LG also open-sourced lightweight and on-device models, maintaining 95% and 86% performance respectively, suitable for smartphones, automobiles, and other industries.
Details link: https://top.aibase.com/tool/exaone-deep
10. Google Chrome Browser to Integrate Gemini AI Assistant, More Convenient Operation!
Against the backdrop of rapid development in internet technology, the Google Chrome browser is about to launch deep integration with the Gemini AI assistant. This feature will greatly enhance the user's online experience, making operation more convenient. Users can directly call the Gemini assistant through the icon on the front end of the window, enjoying support for custom shortcuts and system tray icons, although sidebar fixed mode is not currently supported.
【AiBase Summary:】
✨ The Gemini AI assistant will be deeply integrated into the Chrome browser, enhancing the user's online experience.
🔧 Users can quickly call the Gemini assistant through the icon on the front end of the window, supporting custom shortcuts.
🗣️ The Gemini assistant supports voice search and other functions, but currently does not support sidebar fixed mode.
