AI News 2025-12-17
AI Daily Brief
Summary
Meta released new AI glasses that improve conversation clarity in noisy environments.
OpenAI made GPT Image 1.5 freely available with 4× faster generation speeds.
DoorDash launched AI social app Zesty, which recommends restaurants via a chatbot.
Xiaomi open-sourced a 309B-parameter large model with fast inference and low API pricing.
Apple Music will integrate with ChatGPT, allowing natural-language playlist creation.
Alibaba released the video model Wan 2.6, supporting audio-visual synchronization and role-playing.
Ant Group's AI health app "Ant Afu" saw a surge in downloads, with over 15 million monthly active users and over 5 million daily health queries, becoming China's #1 AI health management app.
Today’s AI News
Meta launches new AI glasses to help you hear conversations more clearly: Meta recently released a new type of AI glasses designed to help users hear people around them more clearly in noisy environments. The glasses use an open-ear speaker design that can amplify the conversation partner’s voice, improving daily communication convenience. At the same time, Meta announced its Messenger desktop app will be discontinued on December 15, 2025. The app had declined in importance due to weaker performance than competitors like Zoom in video call participant limits and screen sharing. Meta has begun gradually integrating Messenger back into the main Facebook app and recommends users switch to the Facebook website or Messenger.com for a better overall experience.
OpenAI late-night surprise: GPT Image 1.5 now free, 4× faster generation: OpenAI officially released the new image model GPT Image 1.5 and made it available to all free ChatGPT users. Compared to the previous generation, the new model achieved four key upgrades: more precise instruction following, finer editing control, better detail preservation, and 4× faster generation speed. It also supports parallel generation of multiple images, and both image input and output prices were reduced by 20%. To lower the barrier, the ChatGPT web interface added a dedicated “Images” entry. OpenAI CEO Sam Altman demonstrated the model’s generated photos on social media, sparking discussion. The move is seen as a key step for OpenAI to drive image generation technology toward mass adoption and build a multimodal ecosystem.
DoorDash launches AI social app Zesty: discover nearby restaurants without reviews: Food delivery platform DoorDash launched a new AI social app called Zesty, aiming to revolutionize how users discover local restaurants. Through an AI chatbot, users can input personalized prompts (e.g., “low-key dinner spots for introverts”) to get precise restaurant recommendations. Zesty aggregates information from DoorDash, Google Maps, TikTok, and other platforms, with the ability to learn user preferences. Beyond AI recommendations, the app includes social features allowing users to share and save recommendations, and follow others’ restaurant updates. Currently, Zesty is in pilot mode and available only in the San Francisco Bay Area and New York.
Xiaomi open-sources 309B-parameter MiMo-V2-Flash large model, inference speed dominates mainstream competitors, API as low as $0.1 per million tokens: Xiaomi recently open-sourced its new foundational language model MiMo-V2-Flash under the MIT license. The model has a total of 309 billion parameters using a sparse activation architecture that activates only 15 billion parameters per inference, maintaining high performance while significantly reducing computational overhead. Benchmarks show its response speed far exceeds mainstream domestic models like Doubao and DeepSeek. To accelerate ecosystem adoption, Xiaomi simultaneously launched a highly competitive API price of just $0.1 per million input tokens. The model targets both developers and plans to integrate into Xiaomi’s “human-vehicle-home full-ecosystem” devices.
Apple Music will integrate with ChatGPT, users can easily create playlists!: OpenAI announced that Apple Music will soon integrate with ChatGPT. Users will be able to ask ChatGPT to create playlists or find music in Apple Music through natural language instructions, such as “create a playlist for my party” or asking about the soundtrack for a specific movie. The integration is expected to launch soon, aiming to provide a more convenient music experience.
Alibaba releases new video model and image model Wan 2.6, supporting audio-visual synchronization and role-playing: Alibaba Tongyi Lab released the Wan 2.6 model, the first domestic video generation model supporting role-playing. Its main features include: native support for audio-visual synchronization with single videos up to 15 seconds; role-playing support that generates consistent, emotionally nuanced character performances based on reference videos and prompts; storyboard control and automatic multi-shot switching to solve the pain point of inconsistent characters and styles when switching shots; and enhanced interaction capabilities for humans, objects, and mixed multi-person scenes. Additionally, its text-to-image capability was significantly improved, with outstanding performance in image accuracy, style consistency, and natural lighting.
Ant Group’s AI health app “Ant Afu” saw a surge in downloads after a brand upgrade and new version release: It has climbed to third place on the overall Apple App Store chart. The app has over 15 million monthly active users and answers over 5 million daily health queries, making it China’s largest AI health management app with a monthly active compound growth rate far above the industry average.