A dark horse enters China's AI race: StartLux(chinaonchina.com)
43 pointsby try-working3 hours ago5 comments
  • ngl9996 minutes ago
    Still censored, no thanks.
  • FailMorean hour ago
    I enjoyed this article... random cat pictures (esp. thumbs up cat picture) kept me going.

    I do think on device AI will be a BIG thing... and basically change how we interact with our computers entirely - you will just ask to get things done. Low latency, free to use as much as you want. I think this future is hard to avoid/this will be a chunk of the future.

    • vkaku6 minutes ago
      I don't think personal computers will be a BIG thing ....

      - IDK, some random mainframe company exec in the 60s ;)

    • MisterTea37 minutes ago
      Like Star Trek -- "tea. earl grey. hot.". Though there will still be tasks where you'd have to likely want or need to pick up a keyboard and mouse. But of course there will be plenty of casual users who dont want or need to do anything interesting and use it for simple tasks like ask for the weather, turn on lights, and brew their earl grey.
    • alansaber24 minutes ago
      For sure, there is a big subset of tasks where ultra low latency >>> other concerns
  • debo_13 minutes ago
    A dark horse named Lux.
  • Karuma3 hours ago
    It's just another Qwen 3.6 27B finetune (not even 3.8...), claiming to beat TB-sized models. This is probably the 57th model like that, just this week, claiming the same thing... And they never deliver, to say the least.
    • dgellowan hour ago
      Didn’t they deliver something here? Why does it matter what base model they used?
      • eightysixfouran hour ago
        What they all deliver is benchmaxxed models that are awful in actual use.
        • vkaku5 minutes ago
          I'm giving out more unpopular opinions today for entertainment and thought :)

          Every model is benchmaxxing today. Many of them are awful until tuned for any particular task.

        • dgellow20 minutes ago
          Qwen is a really awesome model to use!
        • xena44 minutes ago
          But it gives the lab good publicity, as seen in the article.
    • alansaber22 minutes ago
      The finetunes will continue until morale increases.
  • petcat2 hours ago
    The article appears to be slop and even has random model train of thought text interleaved with the article text at various places.

    And you can also tell that this is an obviously Chinese model trained on the Chinese internet in the CCP alternate reality, due to various wild claims:

    > Looks a bit unfamiliar, doesn't it? Don't worry, its founder and CEO is an old acquaintance: Chen Danyan.

    > Known as the "godfather" of programmers in the internet era, his achievements are a matter of public record:

    > Shanda Network's co-founder and the head of Lian Shang Network, one of China's earliest programmers to introduce the concept of "shareware"

    This is a fundamental problem with "open weight" models where you don't know the training data. The biases are real, and they come out in really subtle ways.

    flag

    • BigTTYGothGF2 hours ago
      It's a big world, and I personally don't know anything about the history of software in places other than the west. Could you expand on the particulars of software in China and why those claims are wrong?
    • try-working2 hours ago
      It's a genuine Chinese news article written by a real person. I don't know why they choose to use these types of images, it truly makes it look like slop. I think it's because it's originally a WeChat article posted on their website.
      • Macuyikoan hour ago
        For what it's worth the original article is in Chinese: https://www.qbitai.com/2026/09/483600.html

        The article linked here seems to be a (bad) auto-translated version.

        It's not uncommon for images to be incorporated in an article like this, as they typically get shared around WeChat and correspond better with the 'culture'.

      • petcat2 hours ago
        How do we explain this little blob of sloptext right in the middle:

        > The second task was more straightforward: both models were asked to search for flight tickets in a browser at the same time.

        > I'm unable to open a browser or interact with live websites like Google Flights. I can only process text and don't have real-time browsing capabilities. To find this flight yourself, here's what you'd do: 1. Go to google.com/flights 2. Enter Singapore (SIN) → Beijing 3. Select one-way, set departure date to 5 days from today 4. Filter by "Nonstop" and "Economy" 5. Check the results — note that flights to Beijing may land at either Capital (PEK) or Daxing (PKX). Exclude any Daxing arrivals. 6. Compare prices and pick the cheapest nonstop option landing at PEK. If you'd like, I can help you think through typical price ranges, airline options on this route, or general tips for finding cheap flights.

        > Among them, StartLux-V1.0-27B-Preview completed the search in about 95 seconds, finding an Air China ticket priced at $299.

        • try-working2 hours ago
          they're quoting the models chain of thought.
        • satvikpendem2 hours ago
          Did you even read what it just said above? They're testing the model's capabilities so of course they'd have whatever text the model outputted in the article, that's the entire point.