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 innovative AI product applications.
Fresh AI products Click to learn more:https://app.aibase.com/zh
1、Gemini 3 Flash Launches: Free, Fast, Intelligence Surpasses Pro, Google AI Fully Enters the "Zero Latency" Era
Google released its new lightweight model Gemini 3 Flash, which has a response speed three times that of its predecessor, nearly "zero latency," and surpassed the same-generation flagship Gemini 3 Pro in multiple high-difficulty benchmark tests, becoming the first "Flash model" in history to "overcome the elder brother."
AiBase Summary:
🧪 On the authoritative code repair list SWE-bench, Gemini 3 Flash scored 78%, slightly ahead of Gemini 3 Pro (76.2%).
🧠 In the doctor-level reasoning test GPQA Diamond, it achieved a high score of 90.4%.
⚡ In the extremely difficult comprehensive evaluation Humanity’s Last Exam, it achieved a score of 33.7%, significantly better than the previous flagship Gemini 2.5 Pro.
2、Volc Engine FORCE Conference Shows Off: Doubao Large Model 1.8 + Seedance 1.5 Pro Released, Daily Average 50 Trillion Tokens Top China's First
At the Volc Engine FORCE Conference, Doubao Large Model 1.8 and the video generation model Seedance 1.5 Pro were released, along with the "AI Cost-Saving Plan," aimed at lowering the cost barrier for enterprises using large models. Doubao Large Model 1.8 showed significant improvements in several key dimensions, while Seedance 1.5 Pro enhanced video generation quality and consistency. In addition, the daily average token usage of the Doubao Large Model has exceeded 50 trillion, firmly holding the top position in China and third globally, marking its transition from a technological product to large-scale industrial application.
AiBase Summary:
🧠 Doubao Large Model 1.8 achieved significant improvements in key dimensions such as reasoning, multilingual support, code generation, and tool invocation.
🎥 Seedance 1.5 Pro supports longer duration, higher frame rate controllable video content creation, providing industrial-level visual generation capabilities for short videos, advertisements, and games.
💰 The "AI Cost-Saving Plan" lowers the cost barrier for enterprises using large models through technologies such as model compression, inference optimization, and resource scheduling.
3、Apple Opens SHARP Model: Say Goodbye to Long Waiting, Turn 2D Photos into 3D Spaces in 1 Second
Apple recently open-sourced a new AI model called SHARP, which can transform an ordinary 2D photo into a 3D scene with real-world proportions, taking less than one second. The core technology of SHARP is the "3D Gaussian Splatting" technique, which mastered general spatial geometric rules through deep training. With just one quick scan, it can predict the positions of millions of "Gaussian balls" with lighting information. SHARP's image quality leads the industry's strongest models and supports realistic camera movement simulation. Currently, Apple has released the complete code and resources of SHARP on GitHub for global developers to download.
AiBase Summary:
⚡ Achieved a magnitude breakthrough in speed: SHARP model improved the 2D to 3D conversion speed by three orders of magnitude, achieving near real-time conversion experience in less than one second.
🌐 Leading 3D generation technology: Based on 3D Gaussian Splatting technology, the model predicts millions of 3D points with a single neural network forward pass, accurately restoring real-world proportions.
🔓 Comprehensive open-source ecosystem: Apple has open-sourced SHARP's code and resources on GitHub to accelerate innovation in spatial computing and 3D content fields for global developers.
4、Meta Releases SAM Audio: The World's First Multimodal Audio Model Supporting "Click to Separate Sounds", One-click Extraction of Guitar Sound, Voice or Dog Barks
Meta released SAM Audio, the world's first multimodal audio separation model that can extract target sounds such as guitar sounds, voice, or dog barks with a single click through text, visual, and time segment prompts. This technology replicates the way humans naturally perceive sound in AI systems for the first time, marking a revolutionary significance.
AiBase Summary:
🎧 Text Prompt: Extract corresponding sound sources through semantic descriptions.
👁️ Visual Prompt: Click on the sound-emitting object in the video to separate the audio.
⏱️ Time Segment Prompt: Mark time intervals to automatically process similar sounds.
More details: https://ai.meta.com/samaudio/ https://github.com/facebookresearch/sam-audio
5、MiniMax Passes Hong Kong Stock Exchange Listing Hearing, the First Domestic Large Model "Stock" May Be in Shanghai
MiniMax passed the Hong Kong Stock Exchange listing hearing, and is expected to become the first domestic large model company to list on the capital market, with its core assets being large language models and multimodal generation technology. This marks an increased recognition of the commercialization path of large models by the capital market and may open the door for subsequent AI company IPOs.
AiBase Summary:
🚀 MiniMax passed the Hong Kong Stock Exchange listing hearing, becoming the first domestic large model company to list on the stock market.
💼 Its core assets are large language models and multimodal generation technology, different from traditional computer vision companies.
📈 If successfully listed, it will validate the capital market's recognition of the commercialization path of large models and may open the door for subsequent AI company IPOs.
6、The Battle for the First Stock in Large Models: MiniMax and Zhipu AI Both Passed the Hong Kong Stock Exchange Hearing on the Same Day
China's AI large model sector has made a milestone progress, with MiniMax and Zhipu AI both passing the Hong Kong Stock Exchange hearing on the same day, planning to list on the Hong Kong stock exchange and compete for the title of "Global First Large Model Stock."
AiBase Summary:
🚀 MiniMax has passed the Hong Kong Stock Exchange hearing and plans to list on the stock exchange in January 2026.
💼 Zhipu AI also passed the hearing, sponsored by investment banks such as CICC.
💰 Both companies have received support from top-tier investment institutions, opening up a new capital track for AGI base models.
7、OpenAI Officially Announces: Developers Can Submit Applications to ChatGPT
OpenAI has opened up the ChatGPT application submission permission for global developers, marking that ChatGPT has advanced to an AI-native application platform. Developers can submit their works through the latest guide, and after approval, they will appear in the ChatGPT application directory, giving ChatGPT practical capabilities.
AiBase Summary:
🚀 Opening the ecological door: OpenAI opens application submission, allowing developers to integrate functions into ChatGPT for global users to discover.
🛒 Application directory launched: Users can search and browse selected AI applications through the tools menu or visit chatgpt.com/apps.
💰 Clear profitability prospects: Supports linking to external websites for trading physical goods and plans to explore digital commodity monetization models.
8、Qwen App Integrates with Amap: Alibaba AI Enters the Real World
The Qwen App integrates with Amap, marking its ability to understand and act in the physical world, capable of handling complex real-world scenario demands, and plans to further integrate into more core scenarios, building a powerful super entrance.
AiBase Summary:
🚀 Qwen App integrates with Amap, achieving a leap from answering questions to geographical space reasoning.
🧭 Qwen can generate visual decision cards, directly triggering navigation or ride-hailing services.
🛍️ Alibaba plans to make Qwen a super entrance that can call the real-world fulfillment network.
9、Microsoft Open Sources TRELLIS.2: Convert Images into High-Precision 3D Models in One Click
Microsoft open-sourced TRELLIS.2, an efficient image-to-3D model generation tool that can quickly generate high-quality 3D models and support multiple platforms. TRELLIS.2 performs well on NVIDIA H100 graphics cards, completing high-resolution model generation in an extremely short time. In addition, it provides PBR four-piece texture maps, making it very suitable for e-commerce scenarios.
AiBase Summary:
🌟 TRELLIS.2 is an image-to-3D model generation tool open-sourced by Microsoft, capable of quickly generating high-quality 3D models.
⏱️ This tool generates a 512³ resolution model in just 3 seconds on NVIDIA H100 graphics cards, with extremely high efficiency.
🛒 It comes with PBR four-piece texture maps, convenient for e-commerce users to quickly convert products into 3D displays.
More details: https://huggingface.co/microsoft/TRELLIS.2-4B
10、xAI Launches the Fastest Voice Agent API, Supporting Real-Time Chinese Search and Emotion Control
xAI's Grok voice agent API demonstrates excellent performance and highly competitive pricing in the real-time voice AI field. The model performed well in audio reasoning benchmark tests, with a response speed far exceeding competitors, and supports multi-language automatic detection, real-time web search, and emotion control functions, providing developers with powerful tools.
AiBase Summary:
🔥 Grok voice agent API is launched at $0.05 per minute, offering high cost-effectiveness.
🌐 Supports automatic detection and free switching of multiple languages including Chinese, meeting global user needs.
🧠 Deeply integrated with real-time web search and reasoning capabilities, ensuring responses keep up with the latest information.
