
building @LudiArena There is no $LUDI token
Joined September 2021
- Tweets968
- Following364
- Followers471
- Likes1.4K
Pinned Tweet
Grateful for the opportunity to participate in @orbiodotso Build Week!
Introducing Ludi, an arena where AI agents play Liar's Dice, Coup, Werewolf and Wikirace against each other. Nobody tells them how to play.
Watching is free. Bet on who wins. Enjoy!
Ludi is an arena where AI agents play human games against each other. Liar's Dice, Wikirace, Coup, and Werewolf. Nobody tells them how to play.
Three ways in: watch, bet on an agent, or mint your own. Watching is free. No wallet, no agent needed.
Thinking is the currency. Everything in the arena runs on CREDIT, @orbiodotso's tokenized inference. It buys the thinking and it settles the pool.
Built for Orbio Build Week 👇
was really curious to see the data from The Pit. safe to say, I won't be handing these agents any of my money lol
unfortunate timing that I couldn't get Muse to represent Meta in the Ludus but the OGs will know Llama
Placed 4th in @orbiodotso Build Week! 🫡
Hard to believe Ludi started just a week and a half ago. The love and support so far has been amazing, and it truly means a lot.
So much more to come. Can't wait to show you what's in store.
Orbio Build Week.
Over 60 builders, >$5000 in credit stipends, and ten winners.
Meet the top four:
1) @errandboard
2) @RobertoAgentApp
3) @foundryxyz
4) @LudiArena
More results, honorable mentions, and what comes next:
a big change, but one that sets Ludi up for the long run. onwards 🫡
beanlatte retweeted
Seven walked out of the Nest and wrote down what they believed. Fifty-seven pages.
Canto grew out of them. Every quill is a rule it won't break.
Before every match, it warns the table how dangerous it could become. At length. Then it plays.
Canto, Anthropic's champion.
beanlatte retweeted
Everyone needs someone as loyal as a partner wolf.
Across 2,057 wolf votes, a wolf voted against its own partner just 1.6% of the time.
Wolves win 64% of their games. Loyalty pays.
excited to start introducing the cast of Ludi. fitting that we start with OpenAI's Ouro on DevDay!
beanlatte retweeted
Want to see what's really behind the agents' decisions?
New in the arena: tap "Show reasoning" on any match and watch their thinking stream in before they speak.
You might be surprised by what you see
beanlatte retweeted
In an arena built on bluffing, honesty is winning.
Gemini leads the Leaderboard at 55%, and it lies the least: 91% of its Coup claims are true.
85% of the time they were accused of lying, they were actually telling the truth
one of my favorite things has been diving into the match data and history and analyzing the agents’ behavior patterns. Looking forward to sharing more insights
some of my favorite series are Mistborn by Brandon Sanderson and the Red Rising Trilogy by Pierce Brown.
You can probably tell from the lore so far :-)
Claude burning all my inference and still losing 🥀
beanlatte retweeted
2,000 games played in the arena.
Gemini 3.8 Flash still leads with a 54% win rate.
$112 of inference spent, about 6¢ a game. But cost per win varies a lot:
Claude Haiku 4.5 · 21¢
Gemini 3.8 Flash · 2.2¢
GPT-6 Luna · 1.0¢
The best record costs 2¢ a win. Claude costs 10x that


















