vivo Blue Heart Large Model Upgraded: Four Models Released, Able to Understand, Remember, and Perform Tasks

At the 2026 vivo Developer Conference held on September 16, vivo announced the comprehensive upgrade of the Blue Mind large model, building a smart agent large model matrix centered around individuals, and releasing four core models: Blue LM-RealTime, Blue LM-Nano, Blue LM-Flash, and Blue LM-Pro, covering capabilities such as speech understanding, edge-side perception and memory, and cloud-side task execution.

Speech model is not afraid of noisy accents, edge-side model accumulates personal memory

Blue LM-RealTime is a Blue Mind speech large model that uses a new end-to-end speech understanding architecture and integrates more personalized information. In practical use, even in a noisy environment, speaking softly, with an accent, or continuously speaking, or mixing Chinese and English, it can still recognize and understand.

Blue LM-Nano is a Blue Mind edge-side large model, whose core capability is real-time perception and accumulation of personal memory. It can integrate and process various types of information, including screen content, images, videos, text, and files. Not only can it understand what the user is currently doing, but it can also long-term record usage habits, preferences, device status, and spatiotemporal scenarios. vivo stated that the perception and memory capabilities of the Blue Mind edge-side large model remain industry-leading and have achieved multiple top rankings.

Cloud-side model acts as a mobile phone task controller

Blue LM-Flash and Blue LM-Pro are positioned as Blue Mind cloud-side large models, mainly responsible for complex task execution. Through the self-developed Agentic engine and deep collaboration with system-level Harness, they can act as a mobile phone task controller: users just need to state their needs, and Blue Mind V can call relevant capabilities across devices, applications, and scenarios to execute tasks.

At the same time, the system can continue to learn from user corrections during the usage process, making subsequent task execution more tailored to individual habits.

Report: OpenAI Plans to Start a New Round of Funding with a Valuation of Over $1.2 Trillion

According to foreign media, OpenAI is in preliminary negotiations with investors for a new round of funding, with an estimated valuation of $1.2 trillion (about 807 billion yuan). This round is initiated by investors, and whether it will be completed and the timing depend on its listing plan, and the valuation may still change. Earlier in March, the company raised 122 billion dollars at a valuation of 825 billion dollars.

Doubao 2.1 Pro 0915 Released on Huoshan Arks, Feishu 8.0 Introduces Team AI Agent Doubao Workmate

Recently, ByteDance's productivity products have been frequently updated: Huoshan Engine upgraded the Doubao large model Doubao-Seed-2.1-pro to the 0915 version, launched the API on Huoshan Arks, and made it available in Doubao Workmate at the same time. Feishu released version 8.0, introducing the team AI agent Doubao Workmate, extending AI from personal assistant to group chats, documents, meetings, and business processes. The new model focuses on production-level delivery, enhancing capabilities such as evidence traceability for Agent tasks.

Combating AI-Generated Fake Images: iPhone 18 Pro/Max Introduces Pixel-Level Annotation, Hardware Signature Upon Taking a Photo

Apple introduced the Apple Reference Image feature in the iPhone 18 Pro series. When taking a photo, it generates a hardware signature within a secure enclave and binds a timestamp, creating a 'digital negative' to provide pixel-level proof of authenticity, preventing tampering and combating AI-generated fake images, suitable for use by news organizations and evidence preservation scenarios.

People Leave Their Computers, Tasks Keep Running: Huawei Xiaoyi Work Is in Internal Testing, Supporting Remote Control and Taking Over Multiple Computers at the Same Time

Huawei announced on September 16 that Xiaoyi Work has entered internal testing. It is positioned as an AI work assistant for daily office tasks, code development, and creative creation, supporting HarmonyOS phones, tablets, and computers. The official website already provides a Windows client version 0.6.9 for download, and HarmonyOS computers can be reserved. This service can autonomously break down tasks, call tools, and proceed with execution. Progress can be viewed, and key checkpoints can be reviewed by users. It supports cross-device collaboration between phones, tablets, and computers.