Welcome to the [AI Daily] section! This is your guide for exploring the world of artificial intelligence every day. Every day, we present you with hot topics in the AI field, focusing on developers, so you can understand technical trends and learn about innovative AI product applications.
Fresh AI productsClick to learn more: https://app.aibase.com/zh
1. Xunfei Spark X2.6 is released, with domestic computing power training and agent capabilities being upgraded simultaneously
iFlytek has released the Xunfei X2.6 end-cloud integrated large model, which uses a 293BA30B MoE architecture and has a context length of 1M. The focus is on enhancing the capabilities of code and AI agents. The company has also launched the open-source end-side model X2.64B, and announced limited-time inference prices, further exploring the collaboration between domestic computing power, large models, and end-side devices.
【AiBaseSummary:】
🇨🇳 Based on national-owned computing power training, enhancing code and agent capabilities
🧠 Utilizes the 293BA30B MoE architecture, supporting 1M context
💻 Synchronized release of open-source X2.64B, enhanced edge-cloud collaboration capabilities
2. Hongguo Short Drama launched on “Xiaoguo AI”; short drama platforms begin to adopt conversational interaction.
The Hongguo Short Drama App launched “Xiaoguo AI”. Users can enter the conversation page through the search bar, and also @Xiaoguo AI in the comment section to ask questions. Along with the increasing amount of AI short drama content on the platform, this feature attempts to shift users from one-way viewing to interactive engagement with the content, promoting the short drama platform to explore dialogue and co-creation models.
【AiBaseSummary:】
💬 Both the search bar and the comment section can trigger “Xiao Guo AI”
🎬 The platform synchronously provides multiple AI short drama content
🔄 Short drama consumption has evolved from one-way viewing to interactive experiences
3. iFlytek releases SparkAudio1.0, further integrating voice recognition with emotion understanding
Iluo Xunfei released the SparkAudio 1.0 large model for the Global 1024 Developers Festival. This model integrates capabilities such as speech recognition, machine translation, and semantic understanding into a single framework. It supports speaker separation, environment understanding, and emotion recognition, and has been applied in products like Xunfei Input Method and Xunfei Hear.
【AiBaseSummary:】
🎙️ Release SparkVoice base model SparkAudio1.0
🧠 Combining speech recognition, semantic understanding, and emotion recognition
📱 Multiple iFlytek products have been integrated, with a claimed recognition accuracy of over 96% for the recording loop
4. PHP reveals a high-risk vulnerability in PHP2Shell, and OpenAI works with multiple parties to promote global prevention efforts
OpenAI stated that it participated in identifying and helping fix two high-risk vulnerabilities in the PHP language, with IDs CVE2026109618 and CVE2026109619, and a CVSS v4.0 score of 9.2 for both. The vulnerabilities could lead to unauthorized remote code execution. The PHP official version has already been updated with a fix, and website administrators should check and upgrade as soon as possible.
【AiBaseSummary:】
🚨 Both vulnerability scores reach 9.2, posing a risk of remote code execution
🌐 The PHP ecosystem involves numerous websites, with a wide range of impacts
🛡️ The official patch has been released. It is recommended to upgrade and check the version promptly
5. Yuzhi releases a test advertisement for the H2X humanoid robot, focusing on 8-hour battery life and high-intensity movements
YuShu Technology released a video advertisement for the H2X humanoid robot. The video shows the robot’s dynamic capabilities through scenarios such as hitting car parts, lifting the bumper, and helping children, and emphasizes its up to 8-hour battery life. The video also warns users against imitation and requires them to keep a safe distance of 2 to 3 meters from the robot.
【AiBaseSummary:】
🤖 H2X demonstrates high-intensity movements and dynamic interaction capabilities
🔋 The battery provides up to 8 hours of endurance.
⚠️ The official authorities emphasize safe use, reminding to maintain a safe distance
6. Anthropic launches Claude Duet, and AI agents begin trying to actively perform tasks
Anthropic has launched a new application called Claude Duet, aiming to shift AI from passive response to active task processing. It can aggregate important information from various applications and attempt tasks such as meeting scheduling, financial forecasting, and web application development. The product is currently in a research preview stage, and it will be made available to Claude Pro and Max users in the United States through a waitlist.
【AiBaseSummary:】
🤖 Claude Duet can remain ready in the background
🔔 Actively summarize information and remind users of pending tasks
🧪 Currently in the research preview stage, with limited scope
7. Tesla’s California factory starts mass production of Optimus robots, with net profit declining by 43% in the third quarter
Tesla has started mass production of the Optimus humanoid robot at its Fremont factory in California, with a plan to produce 20,000 units per week by the end of 2027. However, the heavy investment in AI and robotics operations has put pressure on finances, resulting in a 43% decline in net profit for the third quarter to $787 million.
【AiBaseSummary:】
🤖 Optimus humanoid robot officially starts mass production
📈 Musk aims to produce 20,000 units per week by the end of 2027
💰 Net profit in the third quarter decreased by 43% compared to the same period last year
8. Ant Science Technology releases LingDTDecision1.0, reducing token consumption for decision tasks by over 90%
Ant Ant Science released the model LingDTDecision1.0, designed specifically for enterprise decision-making scenarios. It comes in two versions: 4B and 27B, and supports native multiple selections and sequential decision-making. During testing in scenarios such as customer service, risk control, and finance, this model demonstrated advantages in low cost, rapid response, and high accuracy.
【AiBaseSummary:】
🧠 Launched two types of decision models: 4B and 27B
⚡ Token consumption is reduced by up to 90%, and response efficiency improves significantly
🏆 Covers 23 testing items, with the 27B version leading in overall performance