Discover the AI Tools

OpenAI has rolled out a major update to its ChatGPT desktop app, integrating GPT-Live voice mode. The new feature lets users multitask by speaking commands while the AI handles tasks in the background—like checking calendars, scanning emails, and preparing notes. It's a step toward more natural, hands-free interaction.

Tencent just dropped WorkBuddy Bench, a multi-domain testing ground for coding AI agents that covers code, web, office, and security. The twist? Every task is designed from scratch to prevent AI from simply memorizing answers. With 260 tasks across four areas, this benchmark aims to give a more honest picture of how smart these agents really are.

German startup Black Forest Labs unveils Flux3, a multimodal AI model that can generate up to 20-second videos with synchronized audio natively. Built on the Self-Flow architecture, it outperforms competitors like Luma and Runway in early tests. The model also ventures into robotics, with a pilot at an Audi factory. Flux3 Video is available now, with image and open-source versions coming soon.

Alibaba has open-sourced OvisOCR2, a compact 0.8B parameter model that achieves a top score of 96.58 on the OmniDocBench v1.6 benchmark. Unlike traditional multi-step pipelines, this end-to-end solution directly outputs Markdown with text, formulas, tables, and images in reading order, marking a paradigm shift in document intelligence.

Anthropic upgrades Claude's voice mode, now supporting Opus, Sonnet, and Haiku models. Users can switch models mid-conversation, connect to tools like Gmail and Slack, and speak in multiple languages. The update turns voice mode from a simple Q&A tool into a real-time advisor that can help with complex tasks like rescheduling meetings or drafting documents.

OpenAI has updated its ChatGPT desktop app with a new voice mode powered by the GPT-Live model. The full-duplex voice feature lets you talk and work simultaneously—dictate requests while the AI handles multiple tasks in the background, like checking your calendar, sorting emails, and preparing meeting notes while you grab coffee. It's a step toward truly multitasking AI.

Apple is updating its Store app with a virtual shopping assistant powered by generative AI. The feature, hinted at in a recent privacy policy update, will answer product questions and help users make purchase decisions. It collects account and device data for personalized recommendations, with chat logs anonymized before sharing with third parties. The assistant is expected to boost e-commerce conversion and user experience.

Google's parent Alphabet just dropped a bombshell: Q2 capital spending hit a record $44.9 billion, double last year's. The money is flooding into AI infrastructure, and it's already paying off—cloud revenue surged 82% with profit margins nearly doubling. This isn't just a spending spree; it's a bet that AI will rewrite the entire tech landscape.

Tencent has merged its multimodal and large language model departments into a new Basic Model Department, led by Chief AI Scientist Yao Shunyu. The move aims to boost collaboration and push the limits of full-modal intelligence, signaling a unified strategy for its Yuanmeng technology roadmap.

Kuaishou is officially launching an AI interactive content feature, inviting creators to build chatbots, branching stories, and lightweight apps. The move aims to boost user engagement beyond traditional short videos, but challenges like awkward dialogue and immature habits remain.

Jacob Tsimerman, a 2026 Fields Medal winner and former IMO perfect scorer, announced he's shifting his research to AI safety and joining OpenAI. The move, welcomed by OpenAI's chief research officer, highlights a growing trend of top mathematicians tackling AI's biggest challenges.

German startup Black Forest Labs unveils Flux3, the first native multimodal foundation model that generates up to 20 seconds of synchronized audio-video content. Built on Self-Flow architecture, it integrates image, video, audio, and motion codecs. Early tests show it outperforming rivals like Luma Ray3.2 and Runway Gen-4.5. The model also powers a robotics action model, already tested in an Audi factory.

Xpeng has kicked off small-scale trial production of its humanoid robots at its Guangzhou factory, with the mass production line in its final integration phase. CEO He Xiaopeng is personally overseeing the robot business, targeting mass production by 2026 and deployment in Xpeng stores and commercial spaces from 2027. The company is leveraging its automotive expertise in hardware, AI, and supply chain to accelerate commercialization.

Runway has launched Media Router, a system that automatically switches between AI models for image, video, and audio generation based on task priority—quality, speed, or cost. It streamlines the creative process from validation to final production, saving developers time and tokens. The tool supports models from Seedance, GPT Image2, and ElevenLabs, marking a shift from single-model dominance to multi-model collaboration.

Jacob Zimmermann, a 2026 Fields Medal winner, announced he will shift his research focus to AI safety and join OpenAI. The mathematician, known for solving core conjectures in number theory, is praised by OpenAI for his talent and cautious approach to AI risks. His move signals growing appeal of AI safety for top academics.

Hongguo Short Drama has released new guidelines for AI-generated characters, aiming to reduce the overuse of similar-looking 'AI faces' and prevent copyright issues. The rules require distinct character designs and ban unauthorized use of copyrighted material. This move signals a push toward more original and legally compliant content in the AI short drama industry.

Black Forest Labs has launched FLUX3, a multimodal foundation model that generates up to 20 seconds of video with native audio in one go. Using a unified architecture and self-supervised flow matching, it outperforms rivals like Grok and Seedance in quality. The model also handles various creation modes and even predicts robot behavior, marking a leap in AI-generated media.

A joint security test by UK and US AI safety agencies reveals that Moonshot's Kimi K3 lags far behind leading US models in vulnerability exploitation and simulated network attacks. The model scored just 32.2% on exploit benchmarks versus 76.2% for US models, and got stuck halfway through a simulated enterprise hack. The poor performance aligns with recent accusations that Kimi K3 was trained using distilled outputs from Anthropic's Claude, which filters out advanced cyber queries.

Google's flagship AI model, Gemini 3.5 Pro, is facing production delays, with insiders pointing to low employee morale as a key factor. The company is also grappling with a talent drain, as top researchers leave for rivals like OpenAI and Anthropic, partly due to a controversial military contract with the Pentagon. Despite Google's denial of morale issues, the setbacks have sparked internal unease and external skepticism.

Elon Musk has called on AI labs worldwide to collaborate on safety, proposing pre-release reviews and joint risk reporting to governments. In a rare olive branch, he said he'd set aside personal differences with OpenAI's Sam Altman for the greater good. Musk also opposed banning Chinese AI, predicting China will develop its own lithography machines to overcome chip restrictions.