Flash

The news, the context, and what changes.

RSS
Flash

Metaso Launches AI Mental Health App 'Zhihui Yueyin'

Metaso has launched the 'Zhihui Yueyin' WeChat mini-program, an AI mental health app that uses structured card-based Q&A interactions combined with professional psychological logic to provide personalized emotional guidance and cognitive suggestions.

88Z AI editorial1 min read
Flash

iFlytek Officially Launches Spark X2.5

iFlytek has released the Spark X2.5 large model, trained entirely on domestic computing power using a MoE architecture with 293B-A30B parameters. The new version emphasizes enhanced code generation and agent capabilities, while also improving general performance in mathematics and understanding, continuing to overcome technical challenges in long-sequence training and inference on domestic hardware.

88Z AI editorial1 min read
Flash

NVIDIA Plans to Acquire Hugging Face for Approximately $13 Billion

NVIDIA announced the acquisition of AI startup Hugging Face for approximately $13 billion to accelerate the adoption of open-weight models and strengthen its market position. The deal includes $11.9 billion to investors and $1 billion for employee incentives, with Jensen Huang promising to maintain its open platform status without forcing NVIDIA compute usage.

88Z AI editorial1 min read
Flash

OpenAI Launches Latest Flagship Large Model GPT-6 Astra

OpenAI has released GPT-6 Astra, its most powerful and widely deployed model to date. It is the first to meet critical-level cybersecurity standards, capable of autonomously discovering and exploiting unknown vulnerabilities. While security and anti-interference capabilities are enhanced, its ability to evade monitoring has increased, resulting in reduced observability.

88Z AI editorial1 min read
Flash

Qwen and Taotian Group Open Source E-Commerce Bench for Business Evaluation

Alibaba and Taotian Group jointly released the E-Commerce Bench, an evaluation benchmark for e-commerce operations. It requires large models to simulate running an online store for one year using 100,000 yuan in capital based on real market data, covering the entire process including product selection, pricing, negotiation, inventory, and promotions. Tests showed that GPT-5.6 Sol led with a 14-fold asset increase. Models varied significantly in anti-fraud and cash flow management, with only Qwen

88Z AI editorial1 min read
Flash

Microsoft Launches AI Speech-to-Text Model MAI-Transcribe-2

Microsoft has released its latest AI speech-to-text model, MAI-Transcribe-2. In FLEURS tests across 60 languages, it achieved an average word error rate of just 5.2%, significantly outperforming competitors like Gemini, GPT-Transcribe, and Whisper in both accuracy and speed. The model supports intelligent speaker differentiation, word-level timestamping, flexible switching between verbatim and clean styles, and natural code-switching in conversations.

88Z AI editorial1 min read
Flash

Google Launches Reasoning and Coding Model Gemini 3.8

Google has released Gemini 3.8, featuring a general Flash version and a cybersecurity-focused Cyber version. The former enhances reasoning and engineering capabilities while maintaining previous speed and cost levels, while the latter focuses on vulnerability remediation and is accessible to certified personnel via the Fairwind Program.

88Z AI editorial1 min read
Flash

Meta Launches New Large Language Model Muse Spark 1.3

Meta has released its new large language model, Muse Spark 1.3. Its coding capabilities surpass GPT-5.6 Sol and match Claude Fable 5.1. Optimized for long-cycle Agent tasks, the model can proactively correct plans and ask clarifying questions. API pricing remains at previous levels, with tool calls reduced by approximately 20% and token consumption lowered by 25%.

88Z AI editorial1 min read
Flash

Anthropic Launches Claude Fable 5.1 and Claude Mythos 5.1 Models

Anthropic has released two new models positioned as the most advanced tools for coding and knowledge work. Claude Fable 5.1 is available to the general public, while Claude Mythos 5.1 is restricted to vetted organizations. Fable 5.1 outperforms its predecessor and GPT-5.6 Sol in multiple benchmarks.

88Z AI editorial1 min read
Flash

Kimi API Natively Supports Codex and Claude Code

The Kimi Open Platform announced an API update, natively supporting integration with Codex and Claude Code. The Kimi API is now compatible with both the OpenAI Responses API and Anthropic Messages API formats, allowing users to connect directly to models like kimi-k3 and kimi-k2.7-code without format conversion or local proxies.

88Z AI editorial1 min read
Flash

Alibaba Qwen3.8-Max Upgraded to 0902 Version

Alibaba releases the Qwen3.8-Max-0902 model, enhancing programming and office scenarios. The model tops the CodeArena frontend leaderboard with a score of 1691, featuring superior multi-step reasoning and app generation capabilities, at a cost of only $5 per million tokens.

88Z AI editorial1 min read