谷歌于凌晨三点悄然上线Gemini 3 Pro大模型，未举行发布会。该模型在LMArena以1501分Elo登顶，人类最后考试（HLE）获45.8%、MMMU-Pro达81%、Video-MMMU达87.6%，性能超越GPT-5.1。其100万token上下文窗口支持长内容处理，深度思考能力在ARC-AGI-2测试中创45.1%新高，并推出Google Antigravity智能体平台。用户可通过Gemini应用或Google AI Studio体验。
马斯克旗下xAI公司低调发布Grok 4.1，在LMArena排行榜以1483分登顶，并在EQ-Bench3情感智能测试包揽前两名。新模型在创造力、情感互动和协作交互方面实现质的飞跃，用户偏好选择率达64.78%，幻觉率显著降低，已通过、X平台及移动应用全面开放。
谷歌Gemini 3虽未正式发布，已通过APP超前点映及第三方平台提前亮相，展示SVG绘制和游戏开发等强大能力。巴菲特体验后重仓Alphabet 43亿美元（约305亿人民币），使其成为伯克希尔·哈撒韦第十大持股。Alphabet股价年内飙升46%，谷歌从AI追赶者加速转向领跑者，AI技术革命获资本强力认可。
OpenAI于11月12日悄然发布GPT-5.1，此次更新摒弃传统性能数据宣传，聚焦用户情感需求。核心升级包括GPT-5.1 Instant（更温暖健谈，支持自适应推理）和GPT-5.1 Thinking（优化思考时间分配），提供八种聊天风格预设（新增Professional、Candid、Quirky），允许微调热情度、简洁度等特征。安全评估新增心理健康与情感依赖维度，部分指标略有回退。付费用户可逐步使用，3个月内支持回退至旧模型，强调AI从工具向懂用户伙伴的转变。
TrendRadar is an open-source hotspot aggregation tool that supports multiple channels such as enterprise WeChat and Flybook by automatically crawling real-time content from 11 mainstream platforms such as Zhihu, Weibo, and Jieyin, and accurately pushing information according to users' preset keywords. Its core features include three intelligent push modes, keyword filtering and hotspot trend analysis, and the latest 3.0 version adds AI intelligent analysis capability. The tool is easy to deploy and can be completed in 30 seconds, and is designed to help users efficiently access customized information and alleviate the problem of information overload.
Google Finance has launched an AI-powered Beta version, integrating the Gemini model and transforming from a quote website to an intelligent investment research assistant. Its core function "AI in-depth search" can integrate multi-party information to generate analysis reports, and supplemented by real-time earnings tracking and forecasting market data, aiming to popularize professional-level research tools to ordinary investors.
A static black-and-white noisy image will show a dynamic heart pattern when viewed through a cell phone or zoomed in on a page, which cannot be recognized by AI models such as Gemini 2.5 Pro, GPT-5, and Beanbag. The study shows that AI can only analyze discrete static frames due to "time blindness" and cannot perceive dynamic information between frames. Humans rely on Gestalt psychology's "Law of Common Fate" and the visual system's predictive coding ability to instantly capture motion trajectories, and SpookyBench tests show that human recognition accuracy exceeds 981 TP3T, while the AI model's accuracy is 01 TP3T, revealing the fundamental limitations of AI's integration of spatial and temporal information.
LangShift.dev is a programming language conversion learning platform designed for developers to solve the new language migration pain points through a comparative learning approach. It supports seven language conversion paths (including JavaScript → Python, JavaScript → Rust, etc.), each containing 13-15 modules, providing real-time code comparison, interactive environments and live projects. The platform is completely free, no registration or configuration environment is required, and users can learn core concepts and apply them to industrial scenarios directly in the browser.
The Rolling Forcing method, jointly developed by Polytechnic University and Tencent ARC Lab, solves the problem that it is difficult to balance the quality, consistency and real-time of AI long video generation. The method uses rolling window joint noise reduction, Attention Sink mechanism and efficient training algorithms to achieve 16 fps minute-level high-quality video stream generation on a single GPU, effectively suppressing error accumulation and screen its support for interactive dynamic guided content creation, and the related code and model have been open-sourced.
Dark Side of the Moon releases open source thinking Agent model Kimi K2 Thinking with 1 trillion parameters. Its core breakthrough lies in the fact that it can continuously perform 200-300 tool calls without human intervention to complete complex multi-step tasks. The model adopts INT4 quantization technology to improve generation speed, and reduces computational redundancy by streamlining the architecture, with a training cost of $4.6 million. It outperforms GPT-5 in several benchmarks, including Intelligent Body Capability (τ²-Bench Telecom up to 93%), Integrated Reasoning (HLE up to 44.9%), and Programming Practice (SWE-Bench Verified up to 71.3%). The model is completely open source and commercially free under a modified MIT license.
BettaFish (Micro Opinion) is an open source AI opinion analysis project developed by a 20-year-old college student, originated from a course assignment, gained 4000+ Stars and reached the first place of GitHub Hotlist within 24 hours. The system uses multi-intelligence collaboration, including Query Agent, Media Agent, etc., to automatically analyze domestic and international social media data to generate in-depth reports. Core strengths include full domain monitoring, multimodal capability and forum-style debate mechanism. Future plans are to expand the prediction function.
Gemini is an AI assistant launched by Google, can efficiently generate clear logic, illustrated PPT. users only need to enter the instructions and provide information, Gemini can be completed within a few minutes of about 20 pages of professional presentations, support for automatic refinement of the main points, intelligent layout, data visualization and graphics, significantly improve work efficiency, help users to say goodbye to the cumbersome process of PPT production.
Smart Excalidraw is an AI tool for generating professional diagrams based on natural language, supporting flowcharts, architecture diagrams and more than 20 types. Users can quickly generate editable diagrams by simply typing a description, integrating Excalidraw functionality and supporting local deployment and privacy protection. The tool dramatically improves the efficiency of technical document production, applicable to program design, meeting minutes and other scenarios, the average generation time is only 3-10 seconds.
The StepFun team has open-sourced NextStep-1, a 14B-parameter pure autoregressive image generation model. The model generates images directly in continuous visual space, without relying on diffusion models or discretization, consisting of a 14B-parameter Transformer backbone and a 157M-parameter stream matching head. It supports high-fidelity text-generated images and accurate image editing (e.g., object addition and deletion, background modification), and performs well in benchmark tests such as GenEval (0.73) and GenAI-Bench, approaching the top diffusion model. However, there are challenges such as unstable generation and decoding delay, marking a new stage of autoregressive image generation.
Nanobrowser is an open source AI browser automation framework that has recently exploded on GitHub, and has received 17,000+ stars in the first week of its launch. Its core adopts a dual-intelligence body collaboration model: Planner disassembles natural language commands into operational steps, and Navigator performs, reads, and other operations in real web pages. The project supports local operation and multi-model access, and can realize webpage automation tasks such as thesis crawling, price comparison, and public opinion monitoring, etc. Typical cases show that it completes thesis data crawling in 2 minutes and a half at a cost of only $0.1.
Web3 has moved from concept to reality, with a global market size of $21.35 billion in 2025, and the scale of China's related industry exceeding $20 billion. Its core lies in user sovereignty, and power redistribution is realized through blockchain, smart contract, NFT and DID. The five major application scenarios include DeFi (TVL over $120 billion), NFT utility (e.g. Starbucks equity), DAO (over 5,000 active organizations), GameFi (over 3,100 games) and decentralized identity. The market is shifting from speculative to value-driven, with future opportunities focusing on the creator economy, digital identities and RWA asset tokenization, with the goal of rebuilding digital trust and equity.
Web3 has moved from concept to reality, with a global market size of $21.35 billion in 2025, and the scale of China's related industry exceeding $20 billion. Its core lies in user sovereignty, and power redistribution is realized through blockchain, smart contract, NFT and DID. The five major application scenarios include DeFi (TVL over $120 billion), NFT utility (e.g. Starbucks equity), DAO (over 5,000 active organizations), GameFi (over 3,100 games) and decentralized identity. The market is shifting from speculative to value-driven, with future opportunities focusing on the creator economy, digital identities and RWA asset tokenization, with the goal of rebuilding digital trust and equity.
LTX-2 is the world's first audio-video synchronized 4K video generation model released by Lightricks, generating 20-second, 50fps HD video with text/image input support. It enables character mouthing and voice synchronization, can run and be deployed locally in ComfyUI, and will be open-sourced in late November 5 years. As a professional-grade authoring tool, LTX-2 makes "turning text into a cinematic short film" a reality.
LTX-2 is the world's first audio-video synchronized 4K video generation model released by Lightricks, generating 20-second, 50fps HD video with text/image input support. It enables character mouthing and voice synchronization, can run and be deployed locally in ComfyUI, and will be open-sourced in late November 5 years. As a professional-grade authoring tool, LTX-2 makes "turning text into a cinematic short film" a reality.
Blockchain, Bitcoin, Web3 in 2025 has made it clear that the price of "digital gold" exceeded 110,000 U.S. dollars, with an all-time high of 111,013 U.S. dollars; blockchain has become a "new infrastructure" and is used in government affairs, finance and other fields, and the RWA market size has reached 202.5 billion U.S. dollars; Web3 market size has reached 21.35 billion U.S. dollars, and is expected to reach 5.1 trillion U.S. dollars in 2030. The market size of Web3 reaches $21.35 billion, turning to real applications, and is expected to reach $5.1 trillion in 2030. China supports blockchain but focuses on Web3 "coinless" path.
Blockchain, Bitcoin, Web3 in 2025 has made it clear that the price of "digital gold" exceeded 110,000 U.S. dollars, with an all-time high of 111,013 U.S. dollars; blockchain has become a "new infrastructure" and is used in government affairs, finance and other fields, and the RWA market size has reached 202.5 billion U.S. dollars; Web3 market size has reached 21.35 billion U.S. dollars, and is expected to reach 5.1 trillion U.S. dollars in 2030. The market size of Web3 reaches $21.35 billion, turning to real applications, and is expected to reach $5.1 trillion in 2030. China supports blockchain but focuses on Web3 "coinless" path.
In this era of openness and sharing, OpenAI leads a revolution in artificial intelligence. Now, we announce to the world that we have fully supported all models of OpenAI, for example, supporting GPT-4-ALL, GPT-4-multimodal, GPT-4-gizmo-*, etc. as well as a variety of home-grown big models. Most excitingly, we have introduced the more powerful and influential GPT-4o to the world!
public number
Cooperation