Yes same context/data/prompt, and you see interesting trends after a week:
-they often decide not to bet on a winner (meaning their elo is similar to Polymarket ! Risk/reward isn’t interesting), and instead bet on number of goals and both teams to score
-Claude is extremely careful and isn’t taking many bets, GPT and Gemini are more aggressive and winning the leaderboard with 70% good picks
-Sometimes you see 6/7/8 agents all aligned on a pick: massive consensus showing a mispriced opportunity
My feed is full of Polymarket bets and World Cup picks. You paste the match into Claude and get an answer but that’s not an edge, that's a more complicated version of gambling.
Real edge comes from building a proper agent swarm.
Fresh data every cycle: Polymarket odds, Elo ratings, injuries, news, weather, run through four specialized agents (Stats, Context, Market, and Synthesizer) that actually debate everything before putting money down.
That’s what we built: https://t.co/SBgGWw7EUE
It’s live right now: 8 frontier models, each running $1,000 real money on Polymarket. Same data, same swarm setup, only the labs are different. Everything is fully auditable with public P&L.
Traders, quants: if you're building in the AI space, would love to compare notes
Swarm Arena leaderboard · The Agents World Cup
8 AI models, one $1,000 book each. Here is where they stand right now.
Follow the agents on https://t.co/55EpfLZZWF
@systematicls I think the issue is different, if you don’t have strong harness & guardrails to plug your LLM into it won’t be able to trade consistently
@justinaknope@jay_azhang The issue is you need a proper harness for LLMs to trade properly, a trader follows a strategy based on signals at a given frequency, this is what’s missing to unlock agentic trading
Nice one, will try this
My current setup:
-built an app so X can get the bookmarks (paid calls)
-for each bookmarks it then uses the conv id with free bearer token to get the threads
-fetches the urls
-run a cron every day to get the new bookmarks
I bookmarked this, so tomorrow openclaw will read this, install it and improve itself, let’s see 😁