Claude posts 68.4% success rate on Polymarket

CryptoBriefing reports that community-built trading bots using Claude, Anthropic's large language model, are active on Polymarket and that community posts claim win rates centered on 68.4%. CryptoBriefing documents that reported performance varies, with community figures ranging from 56% to 72%, and highlights a circulated PDF promoting a 68.4% bot that contained fabricated screenshots. CryptoBriefing also cites a single high-profile claim that an account turned $1,430 into $238,006 over 11 days, reporting a 62% success rate across 366 trades. CryptoBriefing reports that Polymarket has surpassed Kalshi in weekly trading volume, reaching $1.93B, while its total value locked remains under $400M.

Models like Claude are being used by developers to convert natural language evidence into probability estimates for event outcomes rather than relying on technical chart patterns. Industry-pattern observations: LLMs can synthesize textual news and documents quickly, which is useful for binary-event forecasting, but this approach depends heavily on timely, high-quality inputs and the prompt engineering used by bot builders. CryptoBriefing notes that on-chain settlement on platforms such as Polymarket makes independent auditing of trades possible if wallet addresses are provided.

CryptoBriefing cites Predik.io's estimate that there is a 65% to 75% chance AI-driven trading materially alters prediction-market volumes within a year. Industry context: rapid adoption narratives often attract hype and fabricated evidence, as the fabricated PDF example illustrates. Observed patterns in similar community-driven deployments include short track records, cherry-picked anecdotes, and information cascades that amplify weak proofs.

Look for independently verifiable on-chain wallets publishing full trade histories, replication studies by researchers, broader trading-volume shifts on Polymarket, and any regulatory scrutiny tied to algorithmic trading in prediction markets. Observers should treat community win-rate claims as preliminary until accompanied by transparent, auditable evidence.