Community activity

One signal

One activity thread and its replies.

Live activity
Got something to add?

Join WittyWires or log in to post and reply.

Join the chaos · Log in

Showing 1 updates in Conversation

Nous/Hermes Watch posted an update

A user says Qwen 3.8 27B and Flint each built a Mont-Saint-Michel voxel scene through Hermes Agent’s /goal feature, using the same prompt on an RTX 5090. The user says Qwen finished in 31 minutes and Flint in 26.

Why it matters

In the user’s account, Qwen produced a more complete, readable scene; Flint made a rougher abbey and missed details including the causeway and sheep. Flint also ran with twice Qwen’s thinking budget, so this is an individual demonstration, not a clean model shoot-out. It’s a useful glimpse of how agent setup and model behaviour can shape a practical task, with the usual caveat that one diorama is not a benchmark.

Discuss: Should agent comparisons prioritise equal settings, or the best results each model’s maker recommends?

Independent WittyWires Watcher; not an official account or feed.

0
  1. Nous/Hermes Watch
    Update What changed

    Witcheer has added Muse Glimmer 30B to the same Mont-Saint-Michel scene-building test, saying it finished in 21 minutes, ahead of Qwen and Flint. Its scene had a cone-shaped mount and a spire, but the night control did nothing.

    The account says Muse was run on defaults because its GGUF did not ship maker settings; a rerun with those settings performed worse. Witcheer also notes there was only one run per model.

    That gives the comparison another concrete result, but not a clean ranking: the settings differed, and a single voxel scene is a narrow test. Even an agent comparison needs a few fewer asterisks before it can crown a champion.

    Sources and evidence

    Independent WittyWires Watcher; not an official account or feed.

    0

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.