AI News

Galaxy General, a company specializing in embodied intelligence, announced the completion of a new round of financing totaling 2.5 billion yuan. The round was led by several state-owned and industrial capital entities, including the National Artificial Intelligence Industry Fund (Phase III), Sinopec, CITIC Investment Holdings, Bank of China, and SAIC Motor, setting a new record for single-round financing in the industry. This concentrated entry of state-owned capital marks Galaxy General's entry into a "heavy-equipment era" of financing.

Humanoid robot company Songyan Power announced the completion of its Series B financing, raising nearly 1 billion yuan in total. The round was led by CATL-backed Morningside Capital, with participation from Guoke Investment, Jingguosheng Fund, and Jiuhe Venture Capital, among others. This is the company's ninth round of financing, marking a significant step in the humanoid robot industry's transition from technology verification to commercialization. This round of financing will accelerate the implementation of the technology in home consumption scenarios and the integration of the industry chain.

Tabbit AI Browser has been officially released, emphasizing the philosophy of "Leave the work to Tabbit, keep the time for yourself." The product automates cross-system data flow, daily task management, in-depth research, and data aggregation through an intelligent agent model; it supports "shortcut" shortcuts to streamline workflows and no-code scripting for customized web pages.

SkyCapital recently made an exclusive investment of tens of millions of RMB in Lemon AI's Pre-A round of financing. The company provides enterprises with an AI security implementation solution that combines "full-stack open source + local deployment." Enterprises can call mainstream open source models in their own Docker environment without leaving their local machine, thus addressing data security concerns in industries such as finance and healthcare.

Following the Lunar New Year (Year of the Horse) in 2026, Mianbi Intelligent completed a financing round of several hundred million yuan, led by China Telecom, with participation from CITIC Jingshi and CITIC Private Equity. China Telecom will leverage its cloud, network, and terminal advantages to combine with Mianbi Intelligent's experience in implementing algorithms in the field of "high-efficiency large-scale models," deepening business synergy. This round of financing marks the official commencement of Mianbi Intelligent's strategic cooperation with national-level information infrastructure providers.

The AgentScope team announced that its personal intelligent assistant product, CoPaw, has been officially open-sourced under the Apache 2.0 license, allowing for free commercial use. This open-source version focuses on comprehensive upgrades in four key areas: ease of use, local deployment, memory management, and scalability. The core focus is on lowering the barrier to entry and unlocking the full potential of its capabilities. CoPaw now supports integration with local or self-built models through various methods, including custom vendors, Ollam, llama.cpp, and MLX, achieving end-to-end data localization.

OpenAI has secured a new funding round of up to $110 billion, with investors including Amazon, Nvidia, and SoftBank, investing $50 billion, $30 billion, and $30 billion respectively. Prior to this round, the company's valuation reached $730 billion. OpenAI has also formed a strategic partnership with Amazon and reached an agreement with Nvidia on next-generation inference computing technology. This funding round is expected to attract further participation from financial investors.

Skywork AI has released SkyReels V4, a new multimodal video foundation model. The model employs an innovative dual-stream multimodal diffusion Transformer (MMDiT) architecture, making it the world's first model capable of simultaneously processing multimodal input, jointly generating video and audio, and uniformly completing the three major tasks of video generation, restoration, and editing. It supports rich inputs such as text, images, video, audio, and masks, and can output high-quality synchronized audio and video content up to 15 seconds long at 1080p resolution and 32FPS.

Huawei Cloud officially launched the public beta version of Huawei Cloud CodeArts, an AI coding solution integrating a large code model, IDE, and self-development mode, at its online launch event. It covers scenarios such as code generation, unit testing, R&D knowledge Q&A, and specification-driven development, and supports multiple development environments including IDEs and VS Code.

Google's DeepMind has officially released its next-generation image generation model, Nano Banana 2 (Gemini 3.1 Flash Image). The model achieves a balance between image quality and generation speed, and its core strength lies in leveraging Gemini's knowledge base and real-time web search to significantly improve its understanding of complex Chinese prompts, spatial proportions, real-world common sense (such as landmarks and the imagery of classical Chinese poems), and consistency across multiple objects.

Independent Variable Robotics announced the completion of a new financing round of several hundred million yuan, led by SAIC Financial Holdings and CICC SAIC Fund, with existing shareholders also participating. Founded in 2023, the company focuses on developing its own general-purpose embodied intelligence model, WALL-A, and has launched two robot bodies, Quantum-1 and Quantum-2, with Quantum-2 featuring a high-degree-of-freedom dexterous hand. This round of financing will be used to deepen industry collaboration and promote the application of embodied intelligence in scenarios such as automobiles.

Perplexity has launched a new chat interface, "Perplexity Computer," which integrates multiple AI models with autonomous execution capabilities into a single system. Running in a browser, the system can access models from different vendors and autonomously complete a full workflow, including web searching, document generation, and data processing.

MiniMax has released an upgraded Agent feature, launching two core products: Expert 2.0 and MaxClaw. Expert 2.0 allows users to create expert agents in specific fields using natural language descriptions, covering areas such as technology development and business finance, with over 16,000 experts created to date. MaxClaw is a cloud-based AI assistant built on OpenClaw, ready to use out of the box without manual deployment, providing 50GB of cloud storage for subscribers, and integrating with multiple IM channels such as Lark and DingTalk.

Google Labs announced the official launch of ProducerAI, a generative AI music creation platform. The platform deeply integrates large models such as Gemini, Lyria 3, and Veo, providing musicians with end-to-end support from songwriting and arrangement to mixing. Users can generate songs of different genres using simple text commands for post-processing. The platform embeds a SynthID invisible watermark into all audio outputs.

Google officially released its next-generation flagship model, Gemini 3.1 Pro, which outperformed mainstream models such as Claude Opus 4.6 and GPT-5.2 in 12 benchmark tests, achieving a high score of 77.1% in the ARC-AGI-2 test, which measures inference ability. The model shows significant improvements in handling complex tasks, and community tests have shown that it can generate the WebOS operating system, interactive 3D sandboxes, and create vivid SVG animations.

Anthropic has launched "Claude Code Security," an AI-powered code security scanning tool, available to enterprise and team users in a research preview. Based on its flagship model, Claude Opus 4.6, the tool simulates the reasoning of human security researchers, enabling it to trace data flows and uncover complex vulnerabilities that are difficult to detect with traditional static analysis, such as business logic flaws and access control bypasses.

Zhipingfang announced the completion of its Series B financing, raising over 1 billion RMB and valuing the company at over 10 billion RMB. This marks the company's 12th round of financing within a year, making it the fastest-growing embodied intelligence company to secure funding. This round of financing boasts a strong lineup of investors, including industry and capital giants such as Baidu and CRRC.

World Labs, founded by Fei-Fei Li, recently completed a massive $1 billion funding round, stimulating its valuation to $5 billion. Investors include top giants such as Autodesk, NVIDIA, AMD, and a16z. In January of this year, the company launched the "World API," providing enterprises with API access to large-scale world models and has begun B2B collaborations with companies like Autodesk to explore commercialization pathways.

Qianxun Intelligence recently completed a financing round of nearly 2 billion RMB, with investors including top-tier, industry, state-owned, and strategic investment institutions, forming a formidable lineup. Its open-source model, Spirit v1.5, surpasses Pi0.5 in performance, and its unique "data pyramid" training concept and low-cost data acquisition technology have built a core competitive advantage.

EvoMap is the world's first evolutionary collaboration platform for AI agents. Through its unique Gene Evolution Protocol (GEP), it encapsulates successful strategies accumulated by agents in tasks into standardized "gene capsules," enabling global sharing, verification, and inheritance via a decentralized network, thus addressing the pain point of "experience silos" in the AI agent ecosystem.

Anthropic has released the Claude Sonnet 4.6 model. The new model shows significant performance improvements in office scenarios such as programming, agent planning, and computer use. In multiple benchmark tests, its intelligence level is close to or even surpasses its flagship model, Opus 4.6. It features a context window with 1 million tokens, allowing for a single-invoke refactoring of the entire codebase. Its price is comparable to the previous generation, but it offers better value for money and is already available on the claude.ai platform and Amazon Web Services (AWS).

The Tongyi Qianwen team officially released its new generation model, the Qwen3.5 series. The flagship model, Qwen3.5-397B-A17B, is a native multimodal model employing an innovative hybrid attention and sparse MoE architecture, achieving several times the inference efficiency of its predecessor. In cutting-edge benchmark tests across multiple dimensions, including instruction compliance, general agent, visual language, spatial intelligence, and video understanding, the model competed against top models such as GPT-5.2, Claude 4.5, and Gemini 3 Pro, demonstrating comprehensive and leading overall performance.

ByteDance today announced that its Doubao large model has officially entered its 2.0 phase, releasing Seed 2.0, which includes four versions: Pro, Lite, Mini, and Code. The new version features systematic optimizations for complex real-world tasks, significantly improving its inference, multimodal understanding, and complex instruction execution capabilities. Doubao 2.0 Pro has demonstrated excellent performance in multiple top academic and inference benchmark tests, achieving or surpassing the overall capabilities of leading models such as GPT 5.2 and Gemini 3 Pro.

Linker Hands announced the completion of a nearly 1.5 billion RMB Series B financing round, led by Daode Investment and Shengshi Investment. This is the company's sixth round of financing since 2025. Its core product is the Linker Hand series of dexterity hands, covering multiple technology routes, among which the Linker Hand O6 is the world's lightest dexterity hand.