# CrossOdds vs ArbBets: one checks that it is the same question, the other how many contracts the gap holds for

Two paid Kalshi–Polymarket gap scanners, each doing the half the other skips — and the API that costs 20 USD on one costs 299 on the other.

*https://predictionmarkets.tools/compare/cross-venue-gap-scanners*

- https://predictionmarkets.tools/tools/crossodds.md
- https://predictionmarkets.tools/tools/arbbets.md

## What both of them sell, and what neither can

A Kalshi market and a Polymarket market that ask the same question can sit at different prices, and
when the YES on one plus the NO on the other costs less than the dollar a pair pays out, the
difference is the gap these two products print. Every input is public: both venues' books can be
read without a key. What a scanner sells is the pairing and the arithmetic, and the grid above
cannot show how much of either each one actually does.

The answer is about half each, and the halves do not overlap. That is the finding neither card can
state alone.

The sum of two asks is only the start of what the gap is made of. Neither product's number includes
the capital committed on both venues until each resolves separately, the cost and delay of getting
money out of each, or whether one person may hold both legs at all.
[Why the same contract costs two prices](https://predictionmarkets.tools/guides/why-the-same-contract-costs-two-prices) takes the
gap apart item by item, and [what your money does while you wait](https://predictionmarkets.tools/guides/what-your-money-does-while-you-wait)
covers the part that is time rather than price.

## CrossOdds answers "is it the same question" and not "how big"

**It publishes how a pair is made.** Embedding models pair questions by semantic similarity; a
language model then verifies each pair and attaches a `qualityScore` from 0 to 100 and a `warnings`
list. The API returns both venues' question text separately, and an `arbType` that handles a venue
phrasing the question as the negation of the other. That is enough for a reader to do the check the
score stands in for.

**And its own sample shows why the check is still yours.** The one free pair the home page served on
30 September 2026 matched Kalshi's "more than 10,000 measles cases in 2026" against Polymarket's "at
least 5,000 US cases", at a score of 93.9 and a net spread of 77.87%. Those are two thresholds: a
count between 5,000 and 10,000 loses both legs. The score was high and the pair was two different
contracts. Because both texts were in the row, a reader could see it.

**What it does not return is size.** The API gives top of book only, so a spread printed at the best
quote holds for however many contracts sit at that quote — a number the response does not contain.

## ArbBets answers "how big" and not "is it the same question"

**Its documented arithmetic is about depth.** The API guide walks the full ask side of both books,
binary-searches the largest contract count at which the pair still costs less than the payout after
Kalshi's fee, and turns a dollar budget into a contract count per leg. A row carries both books.
That is the half CrossOdds leaves out, and it decides whether a two-cent gap is worth two cents or
twenty dollars.

**It publishes nothing about matching.** The site says the scanner matches identical events and
monitors 990 or more pairs. There is no method, no score and no field saying whether the two rule
texts, sources or close times were compared. On 250 matched Kalshi–Polymarket events,
[Oddpool](https://predictionmarkets.tools/tools/oddpool) reports that only about a third settled on the same terms. A depth-walked,
fee-netted size is exact arithmetic on a pair that may not be a pair.

## The net spread: one leaves a fee out, the other does not say

ArbBets' own guide computes Kalshi's fee as `0.07 × contracts × price × (1 − price)`, rounded up, and
then states that Polymarket has no trading fee. Polymarket's schedule, as its card and the ArbBets
card both record, charges takers a coefficient of 0.04 to 0.07 on most categories.

Here is what that does to a row, worked from the two cards' formulas. A pair priced at 48 cents for
YES on Kalshi and 49 cents for NO on Polymarket costs 97 cents, a 3-cent gross gap per pair. Kalshi's
taker fee at 48 cents is 1.75 cents, which ArbBets subtracts, printing about 1.25 cents. Polymarket's
fee at 49 cents on a 0.05 coefficient is another 1.25 cents, which it does not. The real figure is
about zero, before withdrawals and before capital time. Near the midpoint, a Polymarket leg the
scanner treats as free can cost as much as the whole margin the scanner shows.

CrossOdds labels its column "after estimated fees" and publishes no formula for either venue, so
the same check cannot be done against it from outside. That is better than a known omission only in
the sense that it may not be one.

**Either way, recompute both legs.** [The fee calculator](https://predictionmarkets.tools/calculators/prediction-market-fees) applies
each venue's published rule and rounding at your price and size, which is the column both of these
should have and neither shows.

## What the API costs per call

Both sell programmatic access, and the units are far enough apart that the comparison is worth
doing, though neither card does it.

- **CrossOdds Pro, 20 USD a month:** up to ten keys at 60 requests a minute, up to 500 rows a call.
  The vendor's own pages say spreads refresh every few minutes, so polling faster buys nothing — but
  the budget never runs out.
- **ArbBets Premium, 299 USD a month:** 5,000 credits a month, and the vendor states the per-call
  price two ways. At 5 credits a call that is 1,000 calls, one every 43 minutes; at 1 credit, 5,000,
  one every 9 minutes. Its dashboard refreshes every three minutes by one of its own pages and every
  second by another. Either way the API cannot keep pace with the dashboard it sits behind.

So the API is fifteen times the price on ArbBets and, at the lower reading of its credits, gives you
a fresh look a little more than once an hour. What it adds for that is the depth walk and two more
venues.

## The rest of the subscription

**Venues.** CrossOdds covers Kalshi and Polymarket. ArbBets adds Opinion and Novig according to its
machine-readable pricing file, though its HTML pricing page names only three. If either extra venue
is where you trade, the choice is made.

**What else you are paying for.** Both bundle screens that are not gap scanners at all. CrossOdds
adds whale trades and a weather screen with a buy direction; ArbBets' Pro tier adds a "+EV" finder,
a weather screen and a "sharp money" scanner. Each of those is a claim about which outcome happens,
not a measurement of a price difference between two venues, and none of it is what this page
compares. The [wallet trackers](https://predictionmarkets.tools/categories/wallet-trackers) category covers what a large-trade
feed shows and cannot show.

**Disclosure.** ArbBets' links to Polymarket carry a referral parameter in the vendor's name, and it
runs its own affiliate programme; neither is disclosed beside the scanner. Its terms speak of odds
data, forbid automated access to the service and treat any data feed as confidential, which sits
oddly beside a tier sold for building bots. CrossOdds' terms name no operator and no governing law,
and its home page carries testimonials quoting monthly returns that nothing verifies.

**Price floor.** CrossOdds has a free tier you can read for a week before paying. ArbBets has none:
the only free material is a delayed sample table and a demo built from one snapshot taken on 19 June
2026, of markets that have since resolved. Its cheapest plan, 59 USD, is the sportsbook product and
contains no prediction-market scanner.

## Which to take

**Start with CrossOdds.** The failure that costs the most in this category is a pair that is not a
pair, and CrossOdds is the one of the two that puts both question texts and a caveat list in the
row so you can catch it. Use the free five a day to find out how often its pairs survive your own
reading of the two rule texts — if they mostly do not, no scanner fixes that. Pay the 20 USD when you
want every pair or the API, and recompute both fees on anything you intend to act on.

**Move to ArbBets Pro, at 149 USD, only when the binding question has become size** — when you
already trust a pair and need to know how many contracts it holds for across the full book — or when
Opinion or Novig is a venue you trade. Before using its numbers, add Polymarket's taker fee back into
every row, and keep reading the rule texts yourself, because ArbBets will not show you which pairs it
was unsure of.

**Pay for ArbBets Premium only for the depth walk delivered as JSON**, and only once you have checked
which of its two credit prices applies to your calls; at one call every 43 minutes, it is a
reconciliation feed rather than a live one. Never buy Basic for this: it is a different product.

If you would rather do the pairing yourself, [matching a question across venues](https://predictionmarkets.tools/how-to/match-a-question-across-venues)
is the procedure and [Oddpool](https://predictionmarkets.tools/tools/oddpool) publishes how few matched pairs settle alike.

## FAQ

### Which is cheaper?

CrossOdds, by a wide margin and on every tier. Its free tier shows five pairs a day and its 20 USD Pro plan includes everything plus the API. ArbBets has no free tier; its 59 USD Basic plan covers sportsbooks only, the prediction-market scanner starts at Pro for 149 USD a month, and the API at Premium for 299.

### Is the spread either one shows net of fees?

Neither shows a figure you can take as final. ArbBets' own API guide nets Kalshi's fee and states that Polymarket charges none, which its card records as wrong for most categories. CrossOdds reports a spread after estimated fees and publishes no formula for either venue. Recompute both legs' fees yourself before reading a row as a margin.

### Which one tells me how much size a gap holds for?

ArbBets. Its documented scanner walks the full ask side of both books and finds the largest contract count at which the pair still costs less than the payout after Kalshi's fee. CrossOdds' API returns top of book only, so the size behind its printed spread is not in the response.

### Which one tells me whether the two markets are the same question?

CrossOdds comes closer. It publishes its matching method — embeddings, then a language model that attaches a 0 to 100 score and a caveat list — and returns both venues' question text. ArbBets publishes no matching method and no score. Neither reads the two rule texts for you, and CrossOdds' own free pair on 30 September 2026 was two different thresholds at a score of 93.9.

### Can I trade the pairs either one shows from the United States?

Only the Kalshi leg. Both scanners return legs on Polymarket's international order book, which does not serve US persons, and neither covers Polymarket US, which is a separate exchange with its own book. Both products are readable from the US; the restriction belongs to the venue.