AIToday
Video GenerationRoboticsOpen-Source AIQiita 機械学習Published: Nov 19, 2026, 16:00 JST

Tencent World Model, Haru 4B, Flash ALOHA Lead AI Roundup

Send on LINE
Tencent World Model, Haru 4B, Flash ALOHA Lead AI Roundup

3 Key Points

  1. What happened

    The roundup spotlights Tencent's commercially usable open-weights World Model, which takes instructions every 5 seconds and stays coherent for about a minute, and Japan's Haru 4B voice model, released roughly three weeks after Block's Desa. It also covers Stanford's Flash ALOHA, which catches balls in mid-air and exceeds 80% success on tasks like folding clothes and battery swapping using ordinary cameras.

  2. Why it matters

    These advances show that open-weight and open-source AI is moving into video-world generation, natural voice conversation, and fast robot learning, putting capabilities once limited to big labs within reach of individual developers. The World Model and Haru 4B being openly available is a key part of that shift.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

The roundup arrives as large language model updates dominate headlines, but the projects collected here show progress across video generation, voice interfaces, 3D content, and robotics. Tencent's World Model stands out for being released as commercially usable open weights, meaning outside developers can build on it directly. Similarly, Haru 4B from Japanese startup Higurashi appeared roughly three weeks after Block's Desa, suggesting a quickening pace in open voice AI research. On the robotics side, Stanford's Flash ALOHA demonstrates that fast, dexterous tasks can be learned from about 50 human demonstrations per task and run with ordinary cameras, while UCLA's PhoneBot shows how low-cost hardware can be paired with smartphone brains and cloud AI. The summary notes that many of these tools are open source or open weight, which could let individual developers prototype robots that move autonomously and converse naturally at home, a capability once confined to large companies and research institutions.

FAQ
What can Tencent's World Model do that earlier video AI cannot?
It can keep generating a world while receiving text instructions or camera moves every 5 seconds, maintaining coherence for up to about one minute without using a game engine.
How fast is Haru 4B and where can I get it?
Haru 4B runs a low-latency loop of about 0.2 seconds on a MacBook and is available on Hugging Face and GitHub.
What tasks can Flash ALOHA perform?
It can catch flying balls, fold clothes, swap batteries, and tidy a table with over 80% success, using only standard cameras.
Qiita 機械学習Read Original Article
Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleKaggle debut: bronze in Kaggriculture at 859th of 10,148