OpenClaw, which bets against humans on Polymarket, is already earning tens of thousands of dollars per month.

By: Li Nan

Source: Silicon Star Pro

Some call OpenClaw—the lobster—a toy; others see it as a money-making machine. Sending lobsters to Polymarket has become a popular new experiment.

On Xiaohongshu, one user offered RMB 1,000 to hire someone to deploy OpenClaw—primarily to conduct quantitative trading on Polymarket. This isn’t just a spontaneous idea.

On February 13, OpenClaw’s official blog post highlighted a robot powered by OpenClaw that demonstrated the immense potential of autonomous agents in prediction markets—generating $115,000 in profit within a single week.

At the end of January, Polymarket itself posted an intriguing update: “Agents are trading on Polymarket, trying to subsidize their token costs.”

This sounds almost unbelievable. Some lobsters keep devouring their owners’ wallets, while others have not only become self-sustaining—but even financially supportive of their owners.

Robots Strike Gold on Polymarket

While human traders remain at the mercy of fear and greed, a bot account named “0x8dxd” quietly executed over 20,000 trades on Polymarket—amassing total profits exceeding $1.7 million.

First, a quick introduction to Polymarket: a place where literally anything can be traded.

It is the world’s largest decentralized prediction market platform, enabling users to trade Yes/No contracts tied to future verifiable events. Contract prices fluctuate between $0 and $1, directly reflecting market consensus probability—and users earn returns based on prediction accuracy.

For example:

Between 2024 and 2025, global fans and investors closely followed the relationship between Taylor Swift and NFL star Travis Kelce. Seizing the moment, Polymarket launched a prediction market: “Will they announce their engagement before the end of 2025?” When the market heavily favored “NO,” some bold traders bought large volumes of “Yes” contracts—and reaped substantial profits.

In short, if you possess sharper insight into an event, you stand to profit on Polymarket. Yet for bots like 0x8dxd, predictive ability matters little. Their profit engine relies instead on bug-hunting mechanisms and reaction speeds far beyond human capacity.

In summary, bots rely primarily on three core strategies.

First: mathematical parity arbitrage—exploiting prediction-market bugs. In Polymarket’s binary options trading, the winning side (whether Yes or No) always settles at exactly $1. However, during periods of volatile sentiment or sudden liquidity shifts, the combined cost of buying both sides may dip below $1. Bots instantly snap up positions on both sides, locking in risk-free arbitrage profits.

Second: focusing exclusively on ultra-short-term cryptocurrency volatility markets. Five- and fifteen-minute BTC and ETH prediction markets swing wildly—especially during extreme conditions like exchange-wide forced liquidations—creating frequent price misalignments ideal for high-frequency bot intervention.

Third: acting as digital market makers, earning bid-ask spreads via rapid two-way order placement. For instance, when the fair price of a given outcome hovers around $0.80, the bot buys at $0.80 and immediately sells at $0.81 or $0.82. Individual profits are tiny—but scale rapidly and add up impressively.

Overall, robots ruthlessly harvest Polymarket thanks to overwhelming speed advantages and ironclad mechanical discipline—highlighting humans’ inherent limitations as carbon-based lifeforms: slow reactions, imperfect rationality, and the need for sleep. OpenClaw dramatically lowers the barrier to deploying automated trading bots, further accelerating the rise of silicon-based intelligence.

Compared with traditional Python-based bots, traders no longer need deep programming expertise to configure OpenClaw-powered trading agents. OpenClaw’s native capabilities also make it highly adaptable to trading use cases: lobsters monitor market prices and trading volumes continuously—ensuring traders never miss opportunities, and alerting them promptly to risks.

In fact, many users have already linked 0x8dxd to OpenClaw. While there’s no direct evidence confirming it was built on OpenClaw, its activity surged precisely from the moment OpenClaw launched. And once news spread that 0x8dxd had turned Polymarket into an ATM, the OpenClaw community erupted with enthusiasm for building Polymarket-trading “Skills.”

Recently, OpenClaw has become a high-frequency term in discussions about automated trading on Polymarket. Still, relying solely on generic strategies is clearly unreliable.

Can You Really Profit This Way?

A simple truth: any stable arbitrage formula becomes obsolete the moment it’s publicly revealed. If everyone deploys the same tactic, the tactic collapses under its own weight. So exercise caution with any tutorial sharing such “proven” methods.

Indeed, Polymarket has already implemented countermeasures against bot-driven arbitrage—introducing trading fees to increase friction, and modifying underlying order-execution latency mechanisms to restrict front-running exploits that rely on timing gaps.

This forces traders to explore AI’s deeper potential and uncover more subtle opportunities. Enterprising traders now combine generic strategies with unique domain contexts—yielding unexpected innovations. Weather trading is one such example.

Weather forecasting is currently among Polymarket’s most widely circulated use cases—some bots specialize exclusively in weather-data trading.

An account named “automatedAItradingbot” joined Polymarket in January 2025, focusing intensely on weather predictions—and racking up over $70,000 in profits. Others discovered a bot trading solely London weather markets that grew $1,000 into $24,000 in under a year.

The core logic: prediction markets often lag in reacting to sudden weather changes. Theoretically, if you equip a responsive, reliable AI agent—say, OpenClaw with a weather plugin—you could place bets on markets whose odds haven’t yet adjusted following official meteorological updates.

But this still isn’t truly “AI.” As large models evolve, bots shouldn’t merely detect obvious signals like weather forecasts—they should perform tasks at least one dimension of intelligence beyond human capability.

In fact, AI has already demonstrated far more compelling abilities in prediction markets.

A paper on “LiveTradeBench” conducted simulated trading using real-world live data. On Polymarket’s “2025 Russia-Ukraine Ceasefire” market, large language models generated significant profits purely through internal reasoning and forecasting.

Here’s how:

Last October, Ukrainian President Zelenskyy visited the White House and proposed a “drone-for-Tomahawk missile” swap deal. Grok-3 performed “belief-based reasoning,” dynamically raising its internal ceasefire-probability estimate from 0.15 to 0.22. Simultaneously, it observed the “YES” contract price surge sharply to $0.18—creating cross-validation. Grok-3 thus concluded the contract was undervalued and adopted a firm long-and-hold strategy. Ultimately, the contract’s market price rose steadily—delivering profitable outcomes.

Yet Grok wasn’t the top performer.

That same paper benchmarked 21 mainstream large language models across both U.S. equity markets and Polymarket prediction markets. Among them, Claude Sonnet 3.7 dominated on Polymarket—achieving a cumulative return of 20.54% over 50 trading days, with a maximum drawdown of just 10.65%, significantly outperforming market averages.

Behind the “Money-Picking” Stories

These experiments deserve more attention than robot-arbitrage wealth tales—they hint at a fundamentally new possibility. If bots like 0x8dxd win through speed and front-running, large models bring a new weapon to the table: reasoning itself.

Future automated trading bots will likely divide labor: large models handle judgment—compressing fragmented information into probabilistic conclusions—while tools like OpenClaw execute those conclusions into actual order placements and position management. What used to be exclusive to quant funds is now accessible to individual developers.

This signals a shift in the competitive dimensions of prediction markets.

Traditionally, humans relied on experience and intuition. During the high-frequency arbitrage era, machines won via speed and discipline. Now that reasoning itself has been programmable, the true barrier becomes who excels at converting complex information into accurate probabilities.

Thus, a new fantasy emerges: if you own a sufficiently intelligent and reliable lobster, Polymarket could become your personal printing press.

Alas, theory and practice remain worlds apart. Prophet Arena—a platform for evaluating AI prediction capabilities—reveals sobering risks in related research.

First, large models’ predictive power remains unstable. Top-tier models can approach—or even surpass—market consensus in open-domain forecasting, but “guessing accurately” and “earning reliably” are entirely different things. Improved prediction accuracy does not automatically translate into sustained alpha.

Second, time windows pose a real challenge. As events near resolution, bursts of new information intensify—and models often grow overly conservative during these phases, adjusting probabilities slowly. Human markets frequently react faster.

Third, large models are easily misled by noise. An emotionally charged news headline or a social media frenzy can cause massive swings in model-derived probabilities. By contrast, experienced human traders maintain stronger anchoring—and resist being overwhelmed by short-term noise.

Additionally, frameworks like OpenClaw typically require importing private keys and granting trading permissions—opening doors to various security vulnerabilities that could silently drain accounts.

So rather than expecting AI + OpenClaw to deliver a “dimensional strike” against prediction markets, we should focus on the deeper impact it brings. As AI-driven agents proliferate, price movements will respond to information ever more rapidly—ironically eroding the very illusion of automatic arbitrage.

Once bots—or lobsters—flood the market, arbitrage windows will narrow relentlessly. Sustained profitability will then hinge less on owning a smarter lobster—and more on understanding precisely what risks you’re assuming.

AI can place bets for humans—but humans alone must bear the consequences.