AI Daily: Moon's Dark Side Opens New AI Framework Kosong; Baidu Releases New Model ERNIE-4.5-VL; GPT-5.1 Makes a Stealth Appearance

Welcome to the "AI Daily" column! This is your guide to exploring the world of artificial intelligence every day. Each 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. Moonshot AI Open Sources New AI Agent Framework Kosong! Asynchronous Orchestration + Plugin Design, Maximizing Developer Flexibility

Moonshot AI has open-sourced the AI agent development framework Kosong. Its core features include asynchronous tool orchestration, plugin-based design, and out-of-the-box Python support, offering developers a more efficient and flexible way to build intelligent agent applications.

【AiBase Highlights:】

🧠 Asynchronous Tool Orchestration Engine Enhances Task Execution Efficiency

🔌 Plugin-Based Design Enables Model Switching Freedom

🚀 Out-of-the-Box Python Framework Accelerates Development

More Details: https://github.com/MoonshotAI/kosong

2. Replit Launches AI Integrations, One-Click Access to 300+ AI Models, Making Building AI Applications Easy

Replit's newly launched AI integration feature greatly simplifies the process of building AI applications. Users can directly access over 300 top AI models without complex setups and quickly complete project builds through an AI assistant.

【AiBase Highlights:】

🤖 Replit AI Integrations allow users to access over 300 AI models without complex configurations.

💡 Smart assistants automatically select appropriate models and handle backend configuration and billing issues.

💰 Usage and costs are transparently displayed on the Replit dashboard for easy cost management.

More Details: https://blog.replit.com/ai-integrations

3. SenseTime Launches SenseNova-SI Model, Performance Exceeds GPT-5 in Spatial Intelligence

SenseTime has made a major breakthrough in spatial intelligence by launching the SenseNova-SI model series, which outperforms international top models in multiple evaluations. The model enhances AI's spatial understanding through systematic training methods and released the EASI evaluation platform to promote standardized technical assessments.

【AiBase Highlights:】

🧠 SenseTime launches the SenseNova-SI model series, significantly enhancing AI's spatial intelligence capabilities.

📊 SenseNova-SI-8B scores 60.99 in spatial intelligence evaluations, surpassing multiple international top models.

🔧 SenseTime uses systematic training methods to enhance the model's spatial understanding through large-scale data augmentation.

More Details: https://github.com/EvolvingLMMs-Lab/EASI

4. Baidu Unveils New Multimodal AI Model ERNIE-4.5-VL

Baidu has released the new multimodal AI model ERNIE-4.5-VL, which not only has strong language processing capabilities but also introduces an "image thinking" function, significantly improving image understanding and processing capabilities.

【AiBase Highlights:】

🧠 ERNIE-4.5-VL model has strong language processing capabilities and introduces an "image thinking" function.

⚙️ The model uses only 3B active parameters, showing excellent computational efficiency and flexibility.

🖼️ "Image Thinking" function supports image zooming and image search, enriching the user experience between images and text.

5. Google Maps Fully AI-Powered! Gemini-Driven "Map Agent" Launched, Generate Interactive Maps with a Single Sentence, Benefiting Both Developers and Users

Google Maps is transforming from a navigation tool into an AI-driven spatial intelligence platform. By integrating the Gemini large model, it introduces three core AI capabilities, providing developers and users with smarter contextual services.

【AiBase Highlights:】

🌍 Builder Agent brings map development into the era of natural language, allowing developers to generate code prototypes simply by inputting instructions.

⚙️ MCP server allows AI assistants to directly access Google Maps technical documentation, lowering the barrier to map development.

🔍 Grounding Lite enables third-party AI models to access map data, achieving high-precision spatial Q&A.

6. AI Teachers Can Now Draw! ChatTutor Launches, Teaching While Drawing, Real-Time Simulations, Teaching Effectiveness Approaching Real Teachers

The article introduces the release of the new generation of visual interactive AI teacher system ChatTutor. This system provides a teaching experience that combines explanation, drawing, and simulation in real-time, covering multiple subjects such as mathematics, physics, logic circuits, programming, and mind maps, significantly enhancing teaching effectiveness.

【AiBase Highlights:】

🧠 ChatTutor enables real-time synchronized whiteboard for teaching while drawing and simulating.

📊 Covers multiple disciplines including mathematics, physics, logic circuits, programming, and mind maps.

💡 It simulates human teachers' heuristic teaching through visual guidance, step-by-step breakdowns, and interactive feedback.

More Details: https://github.com/sheepbox8646/ChatTutor

7. New Version of Google Finance Introduces AI Deep Search and Real-Time Financial Reports, Enhancing Personal Investment Research Experience

The new version of Google Finance introduces powerful AI deep search, predictive market data, and real-time financial report tracking features, enabling ordinary users to understand market dynamics like professional analysts and enhancing personal investment research experiences.

【AiBase Highlights:】

🧠 AI deep search makes complex financial questions simple and understandable.

📊 Predictive market data integrates group wisdom to support investment decisions.

📈 Real-time financial report tracking improves the efficiency of information access for individual investors.

8. GPT-5.1 "Disappears"? OpenAI Quietly Launches Polaris Alpha, 256K Context + Obscure Knowledge Ignites Community Speculation

The article introduces OpenAI's large model named Polaris Alpha, quietly launched on the third-party platform OpenRouter. Its capabilities closely match the rumored GPT-5.1, causing widespread attention. The model supports a 256K context window and knowledge cutoff at October 2024, performing well, especially in long document understanding and creative writing. Additionally, it may be preparing for NSFW features, sparking discussions on commercialization and compliance.

【AiBase Highlights:】

🚀 Polaris Alpha supports a maximum 256K context window, with a knowledge base updated to October 2024, outperforming GPT-4Turbo.

🧠 The model performs well in processing high-density texts such as academic papers and technical manuals, showing strong information extraction and logical summarization abilities.

🎨 In creative writing, Polaris Alpha can generate distinct, emotionally rich multi-version copy, with language naturalness close to humans.