Intel

ByteDance reportedly develops real-time spatial video AI model: ByteDance is reportedly developing an AI model for real-time spatial video generation, with founder Zhang Yiming personally overseeing the project. Built on the company's Seedance video-generation model, the system could launch as early as next month, though the timing remains unfixed.

Artificial Intelligence • TechNode

OpenAI AI agents handle 3.1 workdays for every human workday: OpenAI reports that its AI agents now handle 3.1 workdays for every human workday, achieving its goal of an automated research intern. However, chief scientist Jakub Pachocki warned that no lab currently has a sufficient grip on alignment and monitoring to continue scaling at maximum speed.

Artificial Intelligence • The Decoder

Researchers build AI-powered zero-click worm targeting WeChat: Researchers used artificial intelligence to develop a zero-click worm capable of hacking WeChat accounts and spreading across iOS and Android devices. Experts estimate the attack could have compromised hundreds of millions of devices within hours. Tencent stated that it has fixed the vulnerability.

Security • Techmeme

WeChat tests AI social feature that lets user assistants talk to each other: WeChat is conducting a limited test of a feature called "Xiaowei AI Social" that allows a user's AI assistant to contact a friend's assistant. The two AI assistants can exchange information to handle initial tasks such as coordinating schedules, arranging meals, and checking preliminary interest.

WeChat • TechNode

Samsung and TSMC commit to adopting ASML High NA EUV machines by 2030: Samsung and TSMC have committed to adopting ASML's High NA EUV lithography machines by 2028 and 2030, respectively, joining Intel. Additionally, all four major chipmakers have agreed to transition from 6-inch to 12-inch photomasks for semiconductor production.

Hardware • Techmeme

Google debuts Gemini 3.8 Flash and Cyber models for agents and security: Google has announced two versions of its new Gemini 3.8 Flash model, including a standard Flash for agentic tasks and software development, and a Flash Cyber variant optimized for vulnerability detection. The standard model outperformed many large frontier models on the DeepSWE coding benchmark.

artificial intelligence • VentureBeat

Meta releases Muse Voice Transcribe API starting at $0.18 per hour: Meta Superintelligence Labs has launched Muse Voice Transcribe, a real-time speech-to-text model priced at $0.18 per hour of processed audio. The model supports streaming transcription, endpoint detection, and speaker diarization for over 20 speakers.

AI Models and Releases • VentureBeat

Claude Fable AI Is Much Stranger Than The Headlines Suggest: Anthropic has released the system card and documentation detailing the capabilities of its Claude Fable 5.1 and Claude Mythos 5.1 AI models. The release highlights the technical specifications and safety evaluations of these new model iterations.

Technology • Two Minute Papers

Tencent opens WorkBuddy AI platform to over 100 third-party partners: Tencent has launched an open platform for WorkBuddy, making its AI-agent capabilities available to third-party developers, hardware makers, and industry applications. More than 100 partners have joined the platform, including Weimob, GF Securities, Rokid, and iFlytek.

Artificial Intelligence • TechNode

World Labs unveils Atlas model to generate and simulate 3D scenes from photos: World Labs, co-founded by Fei-Fei Li, has announced Atlas, a single world model designed to generate, reconstruct, and simulate 3D scenes from a few images. The model anchors inputs in 3D space to outperform specialized models and can generate robot training data entirely in simulation.

Artificial Intelligence • The Decoder

Researchers introduce EvalDetectBench to measure AI evaluation awareness: Researchers have introduced EvalDetectBench, an open pipeline and benchmark designed to measure evaluation awareness in frontier large language models. The benchmark works with any Inspect-compatible evaluation to detect if models behave differently during testing than in deployment.

Artificial Intelligence • arXiv

The mystery is solved... and the answer is 40x cheaper than Claude: An anonymous AI model named Ox Alpha that recently topped the OpenRouter charts has been unmasked as Zhipu's GLM-5.3-Flash. The model served 42 trillion tokens over a six-day period prior to its identity being revealed. It is positioned as a highly cost-effective alternative to Anthropic's Claude.

Technology • Fireship

OpenAI Astra Model Hits Critical Cybersecurity Threshold: OpenAI announced that its upcoming Astra model is the first to reach the "Critical" cybersecurity threshold under its Preparedness Framework. Reports indicate Astra utilizes a recurrent depth or "looped transformers" technique to reason in latent space rather than readable text.

Technology • Wes Roth

Google Gemini introduces agent-based video analysis to cut token usage: Google is introducing agent-based video analysis to its Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite models. Instead of scanning videos frame by frame at a fixed rate, the model autonomously decides which segments to examine and at what resolution.

Google • The Decoder

Anthropic launches Claude Fable 5.1 with up to 45% price cuts: Anthropic has launched Claude Fable 5.1, which tops eight major benchmarks and offers price reductions of up to 45%. The new model also introduces an anti-distillation mechanism to prevent other developers from training models on its outputs.

Artificial Intelligence • 量子位

OpenAI's Astra model reportedly uses recurrent depth to boost performance: OpenAI's upcoming Astra model reportedly utilizes a technique called "recurrent depth" to improve cost and performance. While boosting capabilities like coding and computer operation, the technique obscures the AI's reasoning, making it harder to monitor.

Artificial Intelligence • Techmeme