AI News 2025-12-11
AI Daily Brief
Summary
Alibaba Cloud upgraded Qwen3-TTS with a larger voice library and more natural speech.
Meta’s Zuckerberg reportedly took over the AI commercialization roadmap; a closed-source model “Avocado” is planned for spring 2026.
Intel signed a non-binding LOI to acquire SambaNova to bolster its AI accelerator lineup.
Adobe is integrating ChatGPT into Photoshop/Acrobat/Express so users can edit images and PDFs via natural language.
Google updated “AI Mode” search answers with more embedded source links to address publisher traffic concerns.
The 2025 Chinese Word of the Year candidates were announced.
Alibaba released Qwen3-Omni-Flash, a fully multimodal model with streaming responses and broad language coverage.
Former ByteDance leader Pan Xin joined Meituan to lead multimodal AI, while Tongyi Qwen passed 30M MAU and made key productivity features free.
Today’s AI News
Alibaba Cloud’s Qwen3-TTS gets a major upgrade: Alibaba Cloud upgraded its Qwen3-TTS speech synthesis model, significantly expanding the voice library to 49+ high-quality voices covering different genders, ages, and regional traits. It supports 10 major languages (including Chinese, English, German, and French) plus multiple dialects. Speech naturalness is improved, with adaptive speaking rate and prosody based on text. Developers can access it via the Qwen API.
Meta’s Zuckerberg reportedly takes over AI commercialization; “Avocado” to be closed-source in 2026: Internal confirmation suggests Meta CEO Mark Zuckerberg has taken control of the company’s AI commercialization roadmap. Meta’s next-generation model, “Avocado,” is expected in spring 2026 as a closed-source offering delivered via API and hosted services. The model reportedly incorporates training from third-party open-source work including Google’s Gemma, OpenAI gpt-oss, and Alibaba’s Tongyi Qwen, and will run on newly purchased NVIDIA H100 GPUs. The move signals Meta shifting from “open-source champion” toward monetization via APIs, ads, and cloud services.
Intel signs a non-binding intent to acquire SambaNova: Intel signed a non-binding letter of intent to acquire AI chip startup SambaNova Systems (details undisclosed). SambaNova’s key product is the inference-focused RDU (Reconfigurable Dataflow Unit) accelerator. The market is watching how Intel would integrate SambaNova’s technology with existing AI lines such as Habana Labs’ Gaudi family to strengthen competitiveness in the AI accelerator market.
Adobe integrates ChatGPT for natural-language editing across images and PDFs: Adobe announced a deep integration with ChatGPT so users can edit images and PDF files via natural-language conversation inside Photoshop, Acrobat, and Adobe Express. The integration streamlines workflows without switching apps—handling tasks like background blur, parameter adjustments, compression and format conversion, PDF merging, and poster design—then enables one-click handoff to native Adobe apps for deeper editing.
Google improves “AI Mode” search answers with more source links: Google updated its AI search “AI Mode” to include more embedded source links inside AI-generated results, along with short AI-written link explanations to clarify relevance. The update aims to improve the user experience and address concerns that AI answers may reduce publisher traffic. Google is also piloting with multiple media outlets to explore how AI tools can drive audiences back to news organizations.
“Hanyu Pandian 2025” announces Word of the Year candidates: The Hanyu Pandian 2025 campaign released its annual shortlist. Domestic single-character candidates include “韧” (resilience), “创” (innovation), “融” (integration), “智” (intelligence), and “通” (connectivity). Domestic multi-character candidates include “十五五” (the 15th Five-Year Plan), “DeepSeek” (深度求索), and “九三阅兵” (the Sept. 3 military parade). International candidates include slang such as “草台班子” (a slapdash operation) and terms like “geopolitics.” Final results will be announced on Dec 19.
Alibaba releases Qwen3-Omni-Flash, a fully multimodal foundation model: The Qwen team released Qwen3-Omni-Flash, supporting seamless input across text, images, audio, and video with real-time streaming responses. It can generate both high-quality text and natural speech, supports interaction in 119 text languages, and exposes System Prompt customization for persona and style control. It also improves on reasoning, code generation, and multimodal understanding benchmarks. The model is available via API with pricing at ¥1 per 1M input tokens and ¥3 per 1M output tokens, and has a demo inside Qwen Chat. Alibaba’s roadmap includes a 70B lightweight variant in 2025 Q1, voice cloning in Q2, and a video-driven avatar beta in Q3.
Former ByteDance leader Pan Xin joins Meituan to lead multimodal AI: Pan Xin, former head of the visual-model AI platform at ByteDance, joined Meituan to lead multimodal AI innovation. With experience at Google Brain, Baidu, Tencent, and ByteDance, he will drive multiple multimodal applications including LongCat App. Meituan sees AI as central to competitiveness—using AI for internal productivity, upgrading existing products and services, and investing heavily in in-house LLMs. Since 2025, Meituan has launched multiple models and apps including Nocode (AI coding), LongCat-Flash-Chat (LLM), and LongCat-Video (video generation), plus merchant-facing tools such as “袋鼠参谋” and “智能掌柜.”
Tongyi Qwen passes 30M MAU in 23 days and makes key productivity tools free: Alibaba Tongyi Qwen said its beta surpassed 30M monthly active users in just 23 days and announced four core productivity features will be free: AI PPT (generate and revise full decks via chat), AI Problem Solving (photo input with step-by-step guidance and diagnostics), AI Library (a large multi-scenario document repository with smart mounting and downloads), and AI Writing (draft and format documents directly in chat). Alibaba aims to win high-frequency productivity scenarios through top-tier models and “extreme free” user experience—positioning Tongyi Qwen as a “super gateway” in the AI era.
Comments