Discover the AI Tools

Hey, guess what? Moonshot is quietly cooking up Kimi K3.1, and a leak just spilled the beans. We're talking three reasoning gears—Low, High, Max—plus a million-token context window, Agent mode, and even Swarm multi-agent collaboration. It's like giving your AI a turbo boost and a bigger brain. But can it live up to the hype? We'll find out next month.

当大模型自己出题自己练，生成和训练却各跑各的节奏，算力就这么白白烧掉了。腾讯混元团队把经典的"临界批大小"理论搬进在线强化学习，在PPO和GRPO上找到了那个"甜点区"——调一调批大小和学习率，PPO生成吞吐量最高提升2.29倍，GRPO训练时间直接砍掉29%。

Alibaba DAMO Academy has unveiled DAMO EAGLE, an AI model that detects esophageal cancer from plain CT scans—no endoscopy needed. Validated on 80,000 cases across three countries, it achieved 90% sensitivity and 99.2% specificity. The breakthrough could turn routine lung cancer screenings into a powerful tool for catching esophageal cancer early, when it's most curable.

China's livestream shopping market is nearing 7 trillion yuan, but growth is slowing. As AI anchors cut costs by up to 90%, can they truly replace human hosts? Industry leaders like Li Jiaqi say no, pointing to the irreplaceable trust and emotional connection real people bring. The future may be a blend: AI for efficiency, humans for trust.

Huawei just launched Xiaoyi Work, a 24/7 personal assistant that turns a single sentence into a full PPT, data analysis, or market report. It runs on HarmonyOS phones, tablets, and computers, and can even work in the cloud while you're away. New users get a free one-month trial with 1000 AI points. This marks a shift from simple chatbots to reliable long-task executors.

Zhiyuan just hit a huge milestone: its 20,000th general-purpose robot, the Yanzheng A3 Ultra, rolled off the line and went straight to Chimelong Group. That's 10,000 units in just six months. Meanwhile, 300 robots have started regular work at a new embodied intelligence theme park in Hengqin, doing everything from guiding visitors to putting on shows.

Tencent just dropped a new translation app called Hy Translate, and it's aimed squarely at travelers. Think voice, photo, and even offline translation across 33 languages—plus five minority tongues. It's live in the US, Japan, South Korea, and Southeast Asia. Could this be your next must-have travel buddy?

At Meta Connect, Zuckerberg unveiled a new lineup of AI glasses and a lightweight VR headset, all powered by the Muse AI assistant. The star? A camera-free version for privacy buffs. With Muse now in glasses and even getting its own email, Meta is betting big on AI hardware. But will it succeed where others flopped?

OpenAI just rolled out a voice agent for ChatGPT's mobile app, letting you write documents, summarize emails, and even build websites by speaking. Pro and Plus users get the full toolkit, while free users can tap into plugins and connected apps. It's a big shift from Q&A to actually getting things done.

A leak suggests Moonshot AI's next flagship, Kimi K3.1, could arrive as early as October 2026. Internal files point to three reasoning intensities (Low, High, Max), a million-token context, and a Swarm multi-agent mode. The move hints at a shift from chat-focused models to autonomous workhorses.

智能眼镜终于不用时刻联网了。高通和PrismML搞了个1-bit量化视觉模型Bonsai，塞进骁龙AR1平台，离线就能认东西、答问题。内存只占传统4-bit模型的四分之一，速度还翻倍。这意味着更轻的眼镜、更长的续航，还有更少的发热——对天天戴眼镜的人来说，这可不是小进步。

Google DeepMind's Demis Hassabis just confirmed that Gemini 4 is in post-training and might launch well before the end of the year. The company skipped Gemini 3.5 Pro to focus on this next flagship, which is already being tested in internal coding tools. Hassabis says he's '100% confident' in the team, even as rivals race ahead. But is AGI the real goal? He thinks we're asking the wrong question.

DeepSeek年化收入突破10亿美元，较几个月前翻倍。CEO梁文锋在投资者会议上透露，公司计划融资500亿元人民币，并寻求在上海证券交易所上市，投后估值或达5000亿元。这一动作被视为国内大模型商业化的重要里程碑。

Huawei just dropped Xiao Yi Work, an AI assistant that handles office tasks, coding, and creative projects. It syncs across HarmonyOS devices, and for the first 31 days, you get 1000 AI points free. Could this be your new work buddy?

At the UN, OpenAI, Anthropic, and Hugging Face called for international AI safety standards, but the Trump administration rejected any global governance. The clash highlights a growing divide over how to manage frontier AI risks.

Tencent Hunyuan has launched the Tencent Hy Translation app, supporting 33 languages and five ethnic minority languages. It offers voice, photo, and offline translation, with availability in 12 countries and regions. The app is built on the open-sourced Hy-MT2 model, which was released in May and shows significant improvements over its predecessor.

Ant's AI arm just dropped a transparency bomb on HuggingFace. They're publishing training summaries for their Ling, Ring, and Ming model families—but only to comply with the EU AI Act. What does this mean for the future of AI development? We break it down.

China's first AI long drama, 'The Later Journey to the West,' just hit Hunan TV's primetime slot—and not a single camera was used. All 60 episodes were generated by Seedance, with costs slashed to tens of thousands of yuan per episode. The 100-person team rewrote the rules of filmmaking, ditching camera logic for single-frame thinking. Is this the future of television?

Ever notice Claude's writing feels off? An Anthropic engineer spills the beans: the model was optimized for math and code, so it learned to write for other AIs—dense, jargon-heavy, and hard for humans to digest. The good news? The latest version is finding its voice again.

Google just dropped two text-to-speech models that let you craft custom voices with simple descriptions—no technical know-how required. The Gemini 3.8 Flash TTS even replicates a voice from just 30 seconds of audio. With support for over 100 languages and 2,000 ready-made voices, this could change how creators bring characters to life.