Flash

The news, the context, and what changes.

RSS
Flash

Zhipu Launches Latest AI Large Model GLM-5.3

Zhipu has released GLM-5.3, significantly enhancing intelligence and coding capabilities through improved post-training. It ranks first among open-source models in multiple benchmarks. The model also demonstrates strong cybersecurity performance, prompting Zhipu to launch the 'Open Source Shield' initiative for vulnerability detection.

88Z AI editorial1 min read
Flash

MiniMax Open-Sources Music Model MiniMax-Music3

MiniMax has released the open-source music model MiniMax-Music3, capable of generating lyricized songs up to five minutes long. Using a hierarchical architecture with an 8B global model and a 0.6B local model, it outputs 32kHz stereo WAV audio while maintaining consistency in theme, rhythm, and vocal identity across full song structures.

88Z AI editorial1 min read
Flash

Xiaohongshu Open-Sources Multimodal MoE Model dots3-note preview

Xiaohongshu's dots model lab has released dots3-note preview, the first open-source version of the dots3 series. With 280B total parameters and 16B activated parameters, it supports a 512K context window and offers multimodal understanding of text, vision, and speech, performing competitively in various benchmarks.

88Z AI editorial1 min read
Flash

DeepSeek-V4-Pro Official Release

DeepSeek has officially released DeepSeek V4 Pro, available on its app, web interface, and API. The new version significantly enhances agent capabilities, showing strong performance in benchmarks like Terminal Bench and DeepSWE. The API now natively supports the OpenAI Responses API format for Codex compatibility, and the thinking mode offers three selectable levels: low, high, and max.

88Z AI editorial1 min read
Flash

DeepSeek Harness Developer Preview Open for Testing

The DeepSeek Harness Developer Preview (v0.1) is now open for global testing and released as open source under the MIT license. Built on a 'everything is a plugin' architecture, it allows flexible combination of models, tools, and skills, offering four operation modes and full tool invocation capabilities.

88Z AI editorial1 min read
Flash

SpaceXAI Launches Latest Flagship Large Model Grok 4.6

SpaceXAI has officially released the Grok 4.6 model, significantly enhancing capabilities in long-duration agent tasks, complex interactions, and visual processing. It supports data research and codebase analysis. In the AA Intelligence Index benchmark, its performance matches that of GPT-5.6 Sol.

88Z AI editorial1 min read
Flash

Xiaohongshu Open-Sources Speech Synthesis Base Model dots.tts

Xiaohongshu's dots team and Shanghai Jiao Tong University's X-LANCE Lab jointly open-sourced dots.tts, a 2-billion-parameter, fully continuous latent space autoregressive TTS base model. The model does not rely on discrete acoustic tokens, achieves SOTA on the zero-shot voice cloning benchmark Seed-TTS-Eval, and supports timbre fine-tuning and speech editing.

88Z AI editorial1 min read
Flash

Memory Feature Launches, Making Meoo Understand You Better Over Time

Miaowu Meoo has launched a memory feature that automatically organizes and saves long-term user information such as personal background and work habits for intelligent retrieval in future conversations. This feature includes two categories: personalized memory visible only to the user and application memory shared within teams.

88Z AI editorial1 min read
Flash

LTX Open-Sources AI Video Generation Foundation Model LTX-2.5

Open-world model company LTX has released LTX-2.5 with native ComfyUI integration. On dual NVIDIA GB200 systems, it generates a 10-second 720P video in just 6.8 seconds, while the API version outputs 1080P video in 23.7 seconds. The new model offers improved pixel fidelity, multi-shot consistency, and diffusion rendering quality, accessible via Hugging Face, ComfyUI, and API.

88Z AI editorial1 min read
Flash

Qwen Team Open-Sources Qwen3.8-2.4T-A95B Model Weights

The Qwen team has open-sourced the Qwen3.8-2.4T-A95B model, marking the first time weights for a Qwen-Max level model have been released. This MoE model features 2.4 trillion total parameters with 95 billion activated per token and natively supports a 256K context window. It focuses on enhancing capabilities in coding, office tasks, scientific research, and long-cycle agents, leading in benchmarks like PaperBench and TerminalBench. The model supports deployment via frameworks such as SGLang and

88Z AI editorial1 min read
Flash

Manus Resumes Operations as an Independent Company

Manus has announced its separation from Meta to resume independent operations. Due to cross-border regulatory reviews, both parties adjusted their structure. Manus requested some users to back up data by August 23 and will restore services on August 25, offering free access during the transition and return incentives.

88Z AI editorial1 min read