Flash

The news, the context, and what changes.

RSS
Flash

Meta and Broadcom Announce Partnership to Invest Over 1GW of Compute Power in Custom AI Chips

Meta has entered a five-year partnership with Broadcom, initially committing over 1GW of compute power for custom AI chips, aiming to provide personal superintelligence to billions. Meta plans to launch four generations of self-developed MTIA chips within two years and has established a Superintelligence Lab led by the founder of Scale AI as Chief AI Officer.

88Z AI editorial1 min read
Flash

ModelBest Launches Professional Document Agent Workspace Lantay

ModelBest has launched Lantay, a professional document agent workspace, now in public beta. Inspired by Cursor's design philosophy, it features a unified layout and Vibedocing interaction paradigm. Supporting 16 file formats and bulk processing, it targets heavy document users in rigorous industries like law and finance.

88Z AI editorial1 min read
Flash

Seedance 2.0 Fully Opens API Services

Volcengine has officially launched the Seedance 2.0 API, supporting multi-modal input and editing for text, images, audio, and video. It establishes portrait copyright safety standards and is widely used in film, advertising, and autonomous driving.

88Z AI editorial1 min read
Flash

MiniMax Agent Update Redesigns How Agents Operate Computers

MiniMax Agent desktop releases two updates: Pocket enables direct computer task execution via IM apps like Feishu and WeChat; Computer Use allows the agent to control mouse and keyboard like a human, managing local software and system settings.

88Z AI editorial1 min read
Flash

MiniMax Open-Sources New Self-Evolving AI Model MiniMax M2.7

MiniMax has officially open-sourced MiniMax M2.7, a new self-evolving AI model deeply iterated by AI itself. It excels in building complex Agent systems and performs well in software engineering and office scenarios. On its first day of release, it achieved compatibility with major domestic and international chip and inference platforms.

88Z AI editorial1 min read
Flash

MiniMax Launches New AI Music Generation Model MiniMax Music 2.6

MiniMax releases Music 2.6 model with new Cover adaptation feature, optimized performance for traditional Chinese instruments, mid-low frequencies, and vocals. First packet latency is under 20 seconds, supporting precise instruction control, with three open-source Music Skills.

88Z AI editorial1 min read
Flash

MiHoYo Co-founder Cai Haoyu's AI Company Launches First AI Video Model LPM 1.0

Anuttacon, the AI company founded by MiHoYo co-founder Cai Haoyu, has officially launched its first video generation model, LPM 1.0. With 17 billion parameters, the model supports real-time generation of AI character videos that can speak, sing, listen, and express emotions. Its key advantage lies in maintaining identity consistency for videos lasting 22 to 48 minutes. Technically, LPM 1.0 utilizes a diffusion Transformer architecture combined with a causal flow generator, effectively addressing

88Z AI editorial1 min read
Flash

Alibaba Tongyi Open-Sources Multimodal Knowledge Base RAG Framework VimRAG

Alibaba Tongyi Lab has released the VimRAG framework, replacing traditional linear context concatenation with dynamic directed acyclic graphs. By integrating branch trial-and-error, visual energy allocation, and graph-guided optimization, it enhances retrieval and reasoning across text, images, and video. Tests on the Qwen3-VL-8B model show an accuracy of 50.1%, significantly outperforming conventional approaches.

88Z AI editorial1 min read
Flash

Jimeng Launches AI-Native Dynamic Narrative Creation Tool Octo

Jimeng AI has launched Octo, an intelligent creative assistant centered on the Vibe Create philosophy, enabling unrestricted free creation. The tool is now available for beta testing on the official website, aiming to help users easily realize ideas through pure conversational experiences.

88Z AI editorial1 min read
Flash

Meta Launches Native Multimodal Large Model Muse Spark

Meta's Superintelligence Lab has released Muse Spark, its first native multimodal model, which achieved a significant score jump in Artificial Analysis benchmarks and drove strong market performance. The model features visual chain-of-thought and multi-agent orchestration capabilities, demonstrating excellence in specific vertical domains and boosting the company's stock price.

88Z AI editorial1 min read
Flash

Mengyou Intelligent ropet Completes Over $10 Million in Series A Financing

AI companion robot company Mengyou Intelligent ropet has announced the completion of its Series A financing, raising over $10 million. Investors include the Beijing Artificial Intelligence Industry Investment Fund and FreeS Fund. The company focuses on desktop AI pets priced at approximately 2,000 yuan, with nearly 20,000 units shipped to date. Data shows a 90-day retention rate of 80% to 90%, with core users interacting for over two hours daily. The product emphasizes building emotional connect

88Z AI editorial1 min read
Flash

ByteDance Launches Native Full-Duplex Voice Large Model Seeduplex

ByteDance's Seed team has released Seeduplex, a full-duplex voice large model now fully deployed on the Doubao App. Using a 'listen-while-speaking' architecture, the model achieves simultaneous listening and speaking, significantly reducing error replies and interruptions while cutting stop-judgment latency. It improves conversation fluency and user satisfaction, marking the first large-scale implementation of full-duplex voice technology.

88Z AI editorial1 min read