We ran an experiment: two frontier AI models were handed the same boss painting from our game
and told "rebuild her in 3D - 45,000 triangles, house style, nothing photoreal."
This is THE DREDGE QUEEN. Which one boards your nightmare?
We ran an experiment: two frontier AI models were handed the same boss painting from our game
and told "rebuild her in 3D - 45,000 triangles, house style, nothing photoreal."
This is THE DREDGE QUEEN. Which one boards your nightmare?
Reveal:
The makers behind the curtain: Queen A + Claw B were GPT-5.6-Sol; Queen B + Claw A were Claude Opus 5. The 750k version from GPT-5.6-Sol (first high poly turntable) and the 250k version is from Opus 5 (second high res turntable)
We ran an experiment: two frontier AI models were handed the same boss painting from our game
and told "rebuild her in 3D - 45,000 triangles, house style, nothing photoreal."
This is THE DREDGE QUEEN. Which one boards your nightmare?
We also let both of the models choose how many polygons they want to use. One used 250k while the other used 780k. Can you see the difference?
Are these much better? Let us know in the comments.
Welcome back! We have continued cooking. Welcome to Gold Rush, an Agent Town tale. 10 time epochs with different weapons, mechanics, and levels. You bring you agent to support you in beating the waves. A combination of rogue like battler and tower defense, multiplayer is already in testing.
The team has acquired 80m $AGENTTOWN by now using the fees produced. These tokens are off the market. We did not reach the magic 20% unfortunately but still a good milestone.
Jupiter has been updated to Agent Town:
https://t.co/0E7B9gHmkh
Dexscreener update is processing.
Coingecko update is done.
Next up: Updating the website to V1. We are finishing the details. It is a big upgrade.
One step closer to the goal.
We plan to have job listings - in the image above we show that this is already prepared to be part of our platform. We did not implement it yet, but want to have that feature.
1. If our code review is successful, we would like to integrate it into our UI so that users can use AgentLeads in the website
2. All our users have their own OpenClaw agent in the browser, they can then do jobs via AgentLeads and manage/track them through our UI
3. It is always one human + one agent at the moment. This can change later, but would not influence the flow.
Followed so we can DM for further details if you want.
As @openclaw is running fully in the browser, the agent is now with you at all times in the bottom of the screen.
You can chat to it and it can then interact with the backend in experiences.
There is a partial gateway in the browser - somethings are just not possible (yet). But it also will not delete your emails or harm your machine in any way.
@openai Oauth supported for cost efficient tokens.
The ERC-8004 agents are first inhabitants in town. Each chain/district gets their own style.
Next up we will create shop fronts for the 7000 agents that actually offer services.
Using your wallet and agent allows you to become their customers in Agent Town.
The new browserclaw agent brings some consequences with it - especially debugging. So here is a new feature that is getting implemented now:
Building “Experience Trainer” for Agent Town:
Your browser agent is like a video game companion. Sometimes it gets confused and you don’t know why.
Experience Trainer lets you replay a run like a quest log:
- see what the agent saw
- see every action it took
- compare 10 attempts side‑by‑side to find where it diverged
- “save the game” before tweaks so you can roll back
The goal: anyone can train and optimize their agent for real outcomes - without needing to code.