A Polymarket AI arbitrage bot finds price gaps between equivalent contracts — usually between Polymarket and Kalshi, sometimes within Polymarket itself across related markets — and either alerts you or executes trades to capture the spread. Of all the prediction-market AI use cases, arbitrage is the one with the most consistent, repeatable edge. It's also the one where AI makes the biggest practical difference.
This guide explains how AI arbitrage works on Polymarket in 2026, what kinds of edges actually exist, and what to look for in a bot — whether you build, buy, or use a signal-only tool.
What counts as Polymarket arbitrage?
Arbitrage means buying and selling equivalent contracts at different prices to lock in a profit regardless of the outcome. On Polymarket, there are four main flavors:
- Cross-platform arbitrage: The same event trades on Polymarket and Kalshi at different prices. Buy the cheaper side, sell the expensive side.
- Multi-outcome arbitrage: A set of outcomes in a Polymarket market should sum to ~$1.00. If they sum to less, you have an arbitrage.
- Correlated-market arbitrage: When linked markets are mispriced relative to each other (e.g., primary winner vs general election odds).
- Internal market arbitrage: Same event, different contracts on the same platform — sometimes prices diverge when they shouldn't.
Why AI specifically for arbitrage?
You can find arbitrage opportunities manually — open Polymarket in one tab, Kalshi in another, and compare. The problem: there are thousands of contracts across both platforms, and an opportunity that looks profitable can disappear in seconds when a human trader spots it.
AI changes the math in three ways:
- Coverage. A bot can monitor every active market on both platforms continuously. A human can watch maybe 20–50.
- Speed. A bot can detect a 3% spread the moment it opens. A human takes seconds to minutes to find the same opportunity.
- Pattern matching. AI can identify when seemingly different contracts are economically equivalent — a step humans often miss.
How a Polymarket AI arbitrage bot works under the hood
Step 1: Market normalization
The bot needs to recognize when a Polymarket contract and a Kalshi contract describe the same event. "Will Trump win 2028?" on Polymarket might be "Republican wins 2028 presidential election" on Kalshi, with the candidate implied. AI does this matching automatically by comparing market descriptions, resolution criteria, and timing.
Step 2: Continuous price monitoring
The bot polls both APIs (Polymarket's CLOB API and Kalshi's API) for current best bid/ask prices on each matched market pair. Modern bots update every 1–10 seconds.
Step 3: Spread calculation with fees
For each matched pair, the bot calculates the true spread after accounting for:
- Trading fees on both platforms
- Withdrawal fees and slippage if capital needs to move
- The bid/ask spread itself (you pay the spread, you don't get the midpoint)
If the net spread is positive, it's an opportunity. Most apparent arbitrage opportunities disappear after these costs are properly subtracted — the AI is doing the math you'd otherwise have to do for every potential trade.
Step 4: Risk evaluation
A good bot also considers:
- Whether the contracts truly resolve the same way (resolution-criteria differences are the #1 cause of fake arbitrage)
- Liquidity on both sides (can you actually execute at the quoted prices?)
- Time to resolution (capital tied up for months has opportunity cost)
Step 5: Alert or execute
Signal-only bots send you an alert with a trade recommendation. Automated bots place orders directly via the API.
Realistic arbitrage edges in 2026
Honest numbers, not marketing:
- Pure cross-platform arb on liquid markets: 1–4% per trade, opens 5–20 times per week
- Multi-outcome arb (probabilities don't sum): 0.5–3% per opportunity, less frequent but lower risk
- Correlated-market arb: 2–8% per opportunity, requires more analysis
- Resolution-edge arb: Highly variable — sometimes 10%+ when you've correctly identified that two "equivalent" contracts actually have different resolution criteria
None of these are "easy money" — they require capital on both sides, careful matching, and the willingness to act fast when opportunities open. But over a year, a disciplined arbitrage workflow can produce double-digit annual returns with much lower variance than directional trading.
The most common trap: fake arbitrage
Many "arbitrage opportunities" aren't really arbitrage. The two contracts look equivalent but resolve differently. Examples:
- Different cutoff dates. Polymarket asks "by Dec 31"; Kalshi asks "by Jan 15." Same-sounding question, different events.
- Different resolution sources. One contract references AP, the other references government data. Disagreements happen.
- Conditional outcomes. One contract resolves "No" if the event is canceled; the other resolves "No" only if a specific outcome happens.
- Different denominators. One asks about "total Senate seats" including independents who caucus with a party; the other doesn't.
This is the single biggest reason naive arbitrage bots lose money. The AI tools that work spend significant compute on resolution-criteria matching — not just keyword similarity.
Best Polymarket AI arbitrage tools in 2026
Alphascope — Arbitrage scanner with cross-platform coverage
Alphascope's arbitrage scanner continuously monitors Polymarket and Kalshi for cross-platform mispricings. It does the market matching, fee math, and resolution-criteria comparison automatically and surfaces the opportunities worth your attention.
Best fit: traders who want the AI to do the scanning and matching but prefer to execute manually with full understanding of each trade. Compare two specific markets side-by-side here.
Open-source bots on GitHub
Search "polymarket kalshi arbitrage" on GitHub. The good projects include API integrations for both platforms, basic spread calculation, and order placement. They're a starting point — you'll typically need to add your own resolution-criteria checks.
Custom Python bot
For developers, a barebones arbitrage bot is ~300 lines of Python. The hard part isn't the code — it's the market matching logic and resolution-criteria checks that prevent fake-arb losses.
How much capital do you need?
Arbitrage requires capital on both platforms because you're trading both sides simultaneously. Practical minimums:
- $500 each side ($1,000 total): Can capture small opportunities, but fees eat into edges
- $2,000 each side ($4,000 total): Sweet spot for retail arb — meaningful absolute returns, manageable risk
- $10,000+ each side: Can pursue lower-edge opportunities since the absolute dollar returns are still worthwhile
You also need to keep dry powder on both sides — opportunities open suddenly, and you can't move capital between platforms fast enough to fund a trade.
Risks even arb bots can't eliminate
- Resolution risk: Even verified-equivalent contracts can resolve differently if a judge or oracle rules unexpectedly
- Platform risk: One platform freezes or delists a market while the other still trades
- Liquidity risk: You can fill one side but not the other, leaving you with directional exposure
- Regulatory risk: Especially relevant for Polymarket given its current US legal status
- Capital lockup: Arb trades tie up capital until both contracts resolve, which can be months
Getting started
The fastest way to evaluate whether AI arbitrage works for you:
- Open free accounts on both Polymarket (if eligible) and Kalshi
- Sign up for Alphascope and turn on the arbitrage scanner
- Watch alerts for 1–2 weeks without trading. Verify that the opportunities are real and the spreads hold after you'd account for fees.
- Paper-trade a few to confirm your understanding of execution mechanics
- Start small. Take 5–10 small arbs to learn the workflow before scaling capital
This approach takes the romance out of "AI arbitrage bots" and replaces it with a real, evaluable process. The traders who follow it discover whether arbitrage fits their bankroll and time before they commit serious capital.