Flash

The news, the context, and what changes.

RSS
Flash

U.S. Gemini Users Can Connect Google Wallet for Personalized Finance Insights

Google announces that Gemini app users in the U.S. can now connect their Google Wallet accounts to the app. This integration enables Gemini to provide more targeted responses and financial advice based on users' actual spending habits, marking a deeper expansion of personal AI assistants into financial management.

88Z AI editorial1 min read
Flash

Serving GLM-5.3 Across Two NVIDIA DGX Stations with vLLM

As frontier-scale models grow, their weights no longer fit into the GPU memory of a single machine. NVIDIA demonstrates how to use the vLLM framework to serve the GLM-5.3 model across two NVIDIA DGX Stations, overcoming hardware limitations for local deployment of ultra-large models.

88Z AI editorial1 min read
Flash

AllenAI Releases XYBench: Focusing on Deep Reasoning Behind the Question

AllenAI has released XYBench on Hugging Face, designed to evaluate large models' ability to solve the "problem behind the question." This benchmark goes beyond surface-level answer accuracy, deeply examining logical deduction and root cause analysis, offering a new dimension for measuring true AI intelligence.

88Z AI editorial1 min read
Flash

Google Launches Gemini 4 Hydrogen, a High-Efficiency Model in the Gemini 4 Series

Google has released Gemini 4 Hydrogen, focusing on high efficiency and low hallucination rates. The model scores 51 on the intelligence index in High mode, surpassing Argon, and reaches 54 in xhigh mode, approaching Claude Sonnet 5.5 and GPT-6.1 Sol. API pricing is set at $1.5 per million input tokens and $7.5 per million output tokens. It offers 2.5 times the speed of Gemini 3.8 Flash with an 8% hallucination rate.

88Z AI editorial1 min read
Flash

DeepSeek Releases All-in-One Application API

DeepSeek has officially launched its all-in-one application API, integrating core capabilities such as web search, deep reasoning, file processing, and safety moderation. Developers can replicate the full app experience via parameter configuration, while underlying functions like voice and OCR remain available for independent calls.

88Z AI editorial1 min read
Flash

iFlytek Launches SparkAudio 1.0 Voice Foundation Model on Open Platform

iFlytek has launched SparkAudio 1.0, its first voice foundation model trained on fully domestic computing power. The model supports multi-language and dialect recognition, offers deep audio understanding capabilities, and achieves significant performance improvements and cost reductions through three technical innovations.

88Z AI editorial1 min read
Flash

Perplexity Open-Sources Decider v1.3 and Flash v1.3, Free on OpenRouter for a Week

Perplexity has open-sourced its latest decision models, Decider v1.3 (~27B parameters) and the distilled Decider Flash v1.3 (~4B parameters). The update introduces a "locate" feature that extracts text spans with character offsets based on instructions, significantly improving accuracy in entity extraction and classification tasks. Supporting text and image inputs with a 256K context window under the Apache 2.0 license, the models are available for free on OpenRouter until November 19, 2026.

88Z AI editorial2 min read
Flash

Anthropic Launches Private Beta for Self-Hosted Claude Managed Agents

Anthropic has announced a private beta for self-hosted Claude Managed Agents, allowing enterprises to deploy the agent harness, sandboxes, and session history in their own cloud or on-premises environments. This solution privatizes the orchestration layer rather than running the model locally, with model requests still directed to Anthropic or customer-chosen providers, addressing security needs for accessing internal code and systems.

88Z AI editorial1 min read
Flash

Looki Launches Multimodal Model Series Loom

Looki has released its self-developed multimodal model series, Loom, focusing on first-person life understanding and long-term memory. The series includes Loom Sense, Loom Moment, and Loom Memory, designed to optimize continuous perception, event segmentation, and cross-time memory retrieval. It significantly reduces inference costs and improves processing efficiency, turning fragmented observations into traceable personal context.

88Z AI editorial1 min read
Flash

Alibaba's HappyShrimp App Officially Launches

Alibaba's AI music platform HappyShrimp has officially launched on iOS and Android. Users can generate complete songs by inputting emotions or stories, support chat-to-song and video scoring, and use the new 'Make Same Style' feature to replicate creative ideas.

88Z AI editorial1 min read
Flash

Tencent Hunyuan Launches New Real-Time Voice Model Hy Realtime

Tencent Hunyuan has released Hy Realtime, a real-time voice model that enhances spoken language understanding, instruction following, and emotional expression. It supports tool calls like web search and integrates with Tencent Music for song requests. The model is live in Yuanbao voice calls and Tencent Customer Service, with API access coming soon via Tencent Cloud.

88Z AI editorial1 min read
Flash

YC-backed Taan Launches v1-preview, a Controllable Music Foundation Model

Y Combinator-backed Taan released Taan v1-preview on November 12, 2026. The model accepts multimodal inputs like humming, MIDI, images, or scripts to generate music with precise timing control. The team reports a 96% instruction success rate, aiming to solve the lack of controllability in current AI music models.

88Z AI editorial1 min read