Observed arrival · 2026-10-26
AgentArk Hub: playable worlds for agent evaluation
A catalog of Unity task environments for evaluating multimodal agents that act by writing code.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- Researchers evaluating multimodal agents in Unity
- Worth noticing
- The page lists 14 tasks on Kaggle Bench and nine evaluated models, alongside four Colab tutorials.
Field notes
Each task is presented as a playable evaluation world, with the site linking to an open-source environment shell and Hugging Face artifacts, datasets, mods, and releases. Four Colab tutorials cover human play, model replay, online evaluation, and RL training; the page separately lists 14 tasks on Kaggle Bench and nine evaluated models. It says verified releases are expanding toward 1,000-plus tasks in 2026, a forward-looking target rather than current catalog size.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
◎NicheUnusually specific use
One card from the complete issue