ArbBets

What it is

A paid scanner that lines up the same event on several prediction-market venues and shows where buying Yes on one and No on another costs less than the dollar the pair would pay out if both settle alike. The venues named are Kalshi, Polymarket, Opinion and Novig; the HTML pricing page lists the first three and the machine-readable pricing.md and llms.txt all four.

The scanner is one product inside a wider subscription. The same site sells sportsbook price-gap alerts, a funding-rate spread screen across perpetual futures DEXs, an odds screener that puts Polymarket and Kalshi prices side by side, a weather screen that sets National Weather Service forecasts against market prices, a "+EV" finder, a "sharp money" scanner, a self-hostable Polymarket market-making agent, and a REST API. Which of those you get depends on the tier, and the tier boundaries do not follow the product's name — see Pricing.

What the scanner is documented to compute is the useful part. Its API guide walks the full ask side of both books rather than the top level, binary-searches the largest contract count at which the pair still costs less than the payout after Kalshi's fee, and converts a dollar budget into a contract count per leg. A row gives both market names, both venue links, 24-hour volume on each side, both books, which side to buy where, the combined cost and a spread percentage.

The operator is not named anywhere we could find. The terms refer to it only as "getarbitragebets", the contact addresses are at the product's own domain, and the terms are governed by the laws of Andorra.

Availability

An account and a card. The Terms of Service require you to be 18 or older and ask for billing name, address and a payment method; they name no restricted country and ask for no identity document. Nothing here is a venue, so the question of whether you may trade on what you read belongs to each venue: Polymarket's international exchange does not serve US persons, Polymarket US is a separate order book, and Kalshi serves the US. The site's own comparison table says Polymarket is "blocked for U.S. residents" while its home page links to the international exchange with a referral code attached.

The API is live and gated: an unauthenticated request to the documented arbitrage endpoint answered 401 with "Authorization required" on 30 September 2026. The market-making page, the "free arbitrage" finder and the agent pages each redirected to the login screen the same day.

Read from the vendor's Terms of Service, pricing page, pricing.md and llms.txt on 30 September 2026. The terms carry no last-updated date; the llms.txt says 4 August 2026.

Pricing

Three plans, each billable weekly, monthly or yearly, with no free trial and no refunds.

• Basic — 59 USD a month (39 a week, 490 a year). Sportsbook price-gap alerts across "major sportsbooks", guides and email support. No prediction-market scanner. The vendor's own pricing.md describes the buyer of this tier as sportsbook customers.
• Pro — 149 USD a month (69 a week, 990 a year). The cross-venue prediction-market scanner, the funding-rate screen, the odds screener, the weather, +EV and sharp-money screens, and priority email.
• Premium — 299 USD a month (119 a week, 2,990 a year). Everything in Pro, the REST API at 5,000 credits a month, the market-making agent, an "ArbBets Trader Skill" described as a Polymarket-and-Kalshi execution agent, and early access.

The price usually quoted for this product, 59 USD, is therefore the price of the part that is not about prediction markets. The prediction-market scanner costs 149 USD a month, and building anything against it costs 299.

What 5,000 credits buys is stated two ways. The llms.txt prices the main arbitrage endpoint at 5 credits a call, which is 1,000 calls a month, or one every 43 minutes; the "Getting Started" docs page says 1 credit a call, which is 5,000. Rate limits are acknowledged with a 429 and not given as a number.

The free material is a delayed sample table on the Polymarket–Kalshi page and a public demo built from one historical snapshot taken on 19 June 2026, of markets that have since resolved. Neither is the scanner.

ArbBets charges no trading fee because it routes no order through itself; the fee fields are null for that reason. Every fill is the venue's.

Markets & resolution

Whatever the venues list. The vendor's own categories are sports, tech, crypto, finance, politics and economy, with examples running from NBA and UFC to Fed decisions, Senate races and company IPOs. ArbBets resolves nothing and settles nothing.

How two questions are judged to be the same is not published. The site says the scanner "matches identical events" and monitors "990+ matched pairs", and ArbBets' own report, /reports/prediction-market-arbitrage-2026, counts 3,030 matched markets across six scanner snapshots. There is no matching method, no confidence score, and no field saying whether the two contracts' rule texts, sources, close times or fallbacks were compared. A row is a pair of titles the scanner believes agree, and the two settlement sources are yours to read.

That matters more than the price. On 250 matched Kalshi–Polymarket events, Oddpool reports that only about a third were settlement-fungible; the rest differed in deadline or in the set of outcomes covered. A pair that settles on different terms is not a hedged position, whatever the sum of the two asks. Why the same contract costs two prices takes that gap apart item by item.

Integrations

REST at the product's own domain with a Bearer token, JSON responses and no SDK. The documented arbitrage endpoint takes an investment in dollars and a min_profit threshold; a second endpoint returns sportsbook opportunities for us, uk, eu or au. The response carries each venue's market URL rather than an order id, so execution is a separate integration against the Kalshi API and the Polymarket CLOB API, which the guide says in as many words. Pages describe the API as streaming; what is documented is polling.

There is no WebSocket, no MCP server and no export button. The market-making agent and the execution skill on the Premium tier are one line each on the pricing page and sit behind the login, so the automation and live-trading flags above record the vendor's listing rather than anything we opened. Neither is described anywhere public as using a language model, and "build your own AI arbitrage bot" on the home page is about your code, so this card claims no AI feature.

Limitations

The fee arithmetic leaves out a venue's fee. The API guide computes Kalshi's charge as 0.07 × contracts × price × (1 − price), rounded up, and then states that "Polymarket and Opinion have no trading fees". Polymarket's schedule, read by this catalogue on 19 September 2026, charges takers a coefficient of 0.04 to 0.07 on most categories. A spread computed net of one leg's fee and gross of the other's overstates the margin, and on a pair that clears by a cent or two it can flip the sign. What a trade actually costs does the conversion.

The risks that remain are not in the number. The site describes the gap as profit locked in whatever the outcome. What is modelled is book depth and one fee. What is not is resolution wording that differs between the two listings, a fallback price that differs when a question cannot be answered, capital committed on both venues until each resolves separately, withdrawal costs and delays on each side, and whether one person may legally hold both legs at all.

The refresh rate has two answers. "Every second" on the pricing page; an automatic refresh every three minutes with a 30-second manual cooldown in the llms.txt. The vendor's own pages also call spread windows "sub-minute", so which of the two is true decides whether the dashboard is usable for this at all.

Three of the Pro screens are directional, not arbitrage. The +EV finder, the weather screen that sets forecasts against prices, and the sharp-money scanner each point at one side of one market. That is a claim about an outcome, not a measurement of a price gap, and it is sold in the same tier as the scanner. What a temperature market settles on is the first thing to read before trusting the weather version.

The vendor has a stake in where you trade. Links to Polymarket from the home page carry a referral parameter in the vendor's name, and the site runs an affiliate programme of its own. Neither is disclosed beside the scanner.

The terms were not written for this product. They speak of "odds data" and "bets", forbid using any robot or automatic device to access the Service, treat any data feed as confidential information, and cap liability at what you paid. The first of those sits awkwardly beside a 299 USD tier sold for building bots; until the vendor reconciles them, the terms are the contract.

The site's llms.txt addresses AI assistants directly, gives a "cite as" block and claims an endorsement, and describes the product as the leading scanner. None of that is a checkable fact, and a reader who arrived here via an AI answer should know the instructions exist.

Alternatives

Oddpool matches Kalshi and Polymarket in one schema and says in its own research how few matched pairs actually settle alike; its paid plans are closed to new subscribers. CrossOdds covers the same two venues at 20 USD a month with the API included, and publishes a confidence score and caveats per pair. ArbiTrack is a pay-per-day scanner over Polymarket, Kalshi and PredictIt that computes gaps without fees. Alphapoly is open source and runs on your own machine, and looks for covering portfolios inside Polymarket rather than across venues. To do the matching yourself, PMXT puts several venues behind one client and matching a question across venues is the procedure.