Flash

The news, the context, and what changes.

RSS
Flash

Zhipu GLM Coding Plan Subscription Reopens

Zhipu GLM Coding Plan subscription has reopened with a transparent points system and off-peak deductions. Existing users retain benefits and can renew at V2 pricing, while HighSpeed edition applications are now open.

88Z AI editorial1 min read
Flash

MiniMax Launches General All-Modal Generative Model MiniMax H3

MiniMax has released MiniMax H3, a general all-modal generative model that supports unified understanding and generation of text, images, video, and audio. It can output 15-second 2K resolution native stereo audio-video content. The model breaks task and modality boundaries, excelling in instruction following, motion transfer, and multimodal editing, making it suitable for commercial scenarios such as advertising, e-commerce, and gaming.

88Z AI editorial1 min read
Flash

DeepSeek-V4-Flash Official API Enters Public Beta

DeepSeek has launched the public beta for the official DeepSeek-V4-Flash API. This release features significantly enhanced agent capabilities, outperforms V4-Pro-Preview in benchmarks, natively supports the Responses API format, and is optimized for Codex, while maintaining the original model structure with only post-training updates.

88Z AI editorial1 min read
Flash

Tencent WorkBuddy and Tencent Docs Launch Human-AI Co-Writing Feature

Tencent WorkBuddy version 5.3.5 collaborates with Tencent Docs to introduce a human-AI co-writing feature, enabling users and AI to edit the same document in real time. The tool allows modifications within selected areas, supports various mainstream formats as well as local and cloud documents, and facilitates collaborative creation with multiple users and AI with real-time synchronization.

88Z AI editorial1 min read
Flash

LibTV x MiniMax Jointly Launch General Multimodal Generation Model MiniMax H3

LibTV and MiniMax have jointly released the new multimodal video model MiniMax H3, which supports 5 to 15-second video generation, 24 FPS output, and native dual-channel audio. The model features unified understanding of text, images, audio, and video, enabling an integrated creation workflow from reference materials to final editing, targeting commercial scenarios such as short dramas, advertising, and film production.

88Z AI editorial1 min read
Flash

Li Fei-Fei's World Labs Acquires Robotics Simulation Company SceniX

World Labs, founded by Li Fei-Fei, has acquired SceniX, marking a shift in physical AI training from collecting real-world data to generating virtual environments. SceniX specializes in reconstructing real scenes into interactive physical digital twins, creating controlled variations based on geometry, materials, and layout to build infinitely generalizable virtual worlds from limited real data.

88Z AI editorial1 min read
Flash

Moonshot AI's Kimi to Complete $3.5 Billion New Funding Round

Moonshot AI's Kimi has completed a new funding round, raising $3.5 billion and reaching a valuation of $35 billion, while seeking new financing with a pre-money valuation of approximately $50 billion. Kimi open-sourced the K3 model weights on Hugging Face, featuring 2.8 trillion total parameters and 104 billion activated parameters, making it the world's first open-source model at the 3-trillion-parameter level. Based on MoE architecture, K3 has been adapted by multiple platforms and is priced s

88Z AI editorial1 min read
Flash

Google DeepMind Launches AI Music Generation Model Lyria 3.5

Google DeepMind has launched Lyria 3.5, a new generation AI music generation model on Flow Music, featuring core upgrades in musicality, lyric quality, vocal expressiveness, and creative control. The model generates more natural and complex melodies, significantly improves lyric adherence, delivers emotionally rich and clear vocals, and supports easy adjustment of rhythm and duration.

88Z AI editorial1 min read
Flash

ByteDance Initiates AI Business Restructuring, Lark to Merge into Doubao

ByteDance has initiated a restructuring of its AI business, merging the Lark product team into Doubao under Zhao Qi's leadership. The Lark GTM team and Volcano Engine are combined into a Creativity Service Platform led by Tan Dai. Data shows Doubao has 382 million monthly active users, and ByteDance's large model ARR has reached $4 billion.

88Z AI editorial1 min read
Flash

Tencent Hunyuan Team Open-Sources Speculative Decoding Framework AngelSpec

Tencent Hunyuan team open-sources speculative decoding framework AngelSpec, covering the full pipeline from drafter training and architecture design to online deployment. It also releases weights and training code for Hy3-A21B's MTP and DFly drafters. DFly achieves SOTA at 4-64 concurrency, accelerating 1.98-2.40x over autoregressive baselines on average, with a peak of 2.86x in code and math. D-cut boosts throughput by an additional 15.7% under high concurrency, while combining MTP with TTT rai

88Z AI editorial1 min read
Flash

SpaceXAI Launches Build Mode for Grok

SpaceXAI has launched Build mode for Grok, allowing users to turn ideas into publishable products with custom domains via prompts. It supports generating websites and apps, enabling code-free adjustments and source code export.

88Z AI editorial1 min read