Flash

The news, the context, and what changes.

RSS
Flash

Bilibili Open-Sources Industrial-Grade Zero-Shot Voice Cloning Model IndexTTS-2.5

Bilibili has open-sourced IndexTTS-2.5, an industrial-grade zero-shot voice cloning model. With only 0.8 billion parameters, it supports cross-language transfer across Chinese, English, Japanese, Spanish, and Arabic. The restructured inference architecture boosts speed by 2.28 times, achieving a real-time factor as low as 0.20. It also offers seamless speed adjustment from 0.5x to 2.0x and provides pronunciation intervention at three granularities: Pinyin, CMU phonemes, and Japanese Kana.

88Z AI editorial1 min read
Flash

Doubao Launches Student Discount Program

Doubao has launched a special offer for college students, providing 2.5 times the free quota and discounted rates for the professional version to assist with complex tasks like research and creation. Users can verify their status via specific commands on the PC client or by linking their Douyin accounts.

88Z AI editorial1 min read
Flash

Microsoft Launches Programming Model MAI-Code-1.1-Flash

Microsoft has released the upgraded programming model MAI-Code-1.1-Flash, featuring significantly enhanced coding capabilities and new native visual functions. The model shows a 22% improvement in Terminal-Bench 2.1 tests and a 15% boost in .NET tasks. Additionally, token generation speed is increased by 25% with a 25% reduction in consumption, while the price is cut to one-quarter of the original. The new model is gradually being integrated into Copilot platforms like VS Code and JetBrains, wit

88Z AI editorial1 min read
Flash

OpenAI Launches AI Cybersecurity Model GPT-5.6-Cyber

OpenAI has expanded its Daybreak cybersecurity defense service by adding Blue and Red tiers. The Red tier includes the new GPT-5.6-Cyber model, trained for security testing, which is currently available only to trusted partners such as Accenture and IBM.

88Z AI editorial1 min read
Flash

Meta Open-Sources 30 Billion Parameter LLM Muse Glimmer

Meta has released Muse Glimmer, an open-source large language model with 30 billion parameters. Optimized for local deployment, it uses 4-bit quantization and DFlash speculative decoding to run smoothly on devices with 24GB VRAM, such as Macs or consumer GPUs, supporting offline multimodal perception, tool use, and code generation.

88Z AI editorial1 min read
Flash

ZCode Fully Upgraded, GLM Coding Plan Quotas Restored for All Users

ZCode, under Zhipu AI, has undergone a comprehensive upgrade, introducing four core features: Goal, Subagents, Remote Control, and Off-Peak Tasks. The Goal mode enables autonomous breakdown and delivery of complex tasks, Subagents facilitate parallel multi-agent collaboration, Remote Control allows management via mobile or WeChat, and Off-Peak Tasks run without points during low-traffic periods. Meanwhile, GLM Coding Plan quotas have been uniformly reset, with all user balances fully restored.

88Z AI editorial1 min read
Flash

Tencent Launches Windows Desktop Customization Tool 'Xiao'e Desktop'

Tencent has launched Xiao'e Desktop, a Windows customization tool offering free, ad-free access to millions of 4K wallpapers. It features a lightweight AI widget for tasks and file search, targeting students and individual users to complement its enterprise-focused WorkBuddy.

88Z AI editorial1 min read
Flash

NVIDIA Open-Sources Autonomous Driving AI Inference Model Alpamayo 2 Super

NVIDIA has released Alpamayo 2 Super, an open-source autonomous driving model built on Cosmos 3 Super Reasoner. Licensed under OpenMDW 1.1 for commercial use and fine-tuning, it ranks top in benchmarks like LingoQA. The model outputs five result types, including trajectory planning and causal reasoning, ensuring transparent and verifiable decision-making.

88Z AI editorial1 min read
Flash

Tencent Hunyuan Team Launches 3D World Generation Framework WorldClaw

Tencent Hunyuan's 3D research team has released WorldClaw, an agent-based framework for generating open 3D worlds. It converts text prompts into explorable and editable 3D environments, maintaining terrain consistency while adding local details, and supports object-level editing and game engine integration.

88Z AI editorial1 min read
Flash

Alibaba Launches New Generation Video Generation Model Wan3.0

Alibaba has opened public testing for its new generation video generation model, Wan3.0. The model can generate 30-second videos in a single run, supporting continuous camera movement and single-take narratives. In addition to text, image, audio, and video inputs, it now supports document formats such as doc, xls, ppt, pdf, and md for the first time, allowing office materials to be directly converted into teaching courseware or product demonstration videos. Portrait generation aims for unique in

88Z AI editorial2 min read