Between 6 July 2026 and 12 August 2026 we recorded the pre-game price Kalshi and Polymarket showed on the same 371 MLB games. The two venues posted an identical number on 212 of them, and the widest gap in the whole window was 3.0 cents. Across every graded game, neither venue's prices were measurably more accurate.
Live prices by book on today’s MLB card: the per book price grid.
What the two venues actually did
Kalshi and Polymarket both run moneyline markets on the same Major League Baseball games. We hold each venue's own price history, so for every game in the window we can line up the last price each venue showed before first pitch and ask a simple question: how far apart were they?
The answer, over 371 graded games from 6 July 2026 to 12 August 2026, is that they were usually not apart at all.
| Gap between the two venues | Games | Share | Mean gap (cents) |
|---|---|---|---|
| identical to the cent | 212 | 57.1% | 0.0 |
| under 1 cent | 9 | 2.4% | 0.4 |
| 1 to 2 cents | 144 | 38.8% | 1.0 |
| 2 to 5 cents | 6 | 1.6% | 2.2 |
| 5 cents or more | 0 | 0.0% | n/a |
Gap is the absolute difference between the two venues' last pre-game prices on the away team. The widest gap on any single game in the whole window was 3.0 cents; the mean was 0.4 cents.
That is the first finding, and it is the one that matters most. Two separate exchanges, with separate users and separate order books, priced the same 371 baseball games and landed on the same number more than half the time. Not one game in the window saw them five cents apart.
When they did disagree, which price was closer?
Answering this requires holding both venues' prices from before each game and then checking them against what happened. For each game we take the distance from each venue's price to the result, one for a win by the away team and zero for a loss, and record which venue was nearer.
| Disagreement size | Games | Kalshi closer | Polymarket closer | Kalshi share | Binomial p |
|---|---|---|---|---|---|
| any disagreement | 159 | 92 | 67 | 57.9% | 0.0567 |
| half a cent or more | 154 | 88 | 66 | 57.1% | 0.0903 |
| 1 cent or more | 150 | 86 | 64 | 57.3% | 0.0861 |
| 2 cents or more | 6 | 5 | 1 | 83.3% | 0.2188 |
Binomial p is an exact two-sided test against a 50/50 split, computed in the same committed script that produced the counts. Games where the two venues showed the identical price are excluded from this table, because neither price can be closer than the other.
Across the 159 games where the two venues differed at all, Kalshi's price was closer to the result 92 times and Polymarket's 67 times. Four things have to be said about that number before it can mean anything:
- The disagreements are tiny. The mean gap between the two venues was 0.4 cents and the largest was 3.0 cents. Being closer on a one cent difference is a coin-flip sized question on any individual game.
- Overall accuracy is a tie. Across all 371 graded games the Brier score was 0.2415 for Kalshi and 0.2421 for Polymarket, a difference of 0.0006. At this sample size that difference is noise.
- We looked at four thresholds. The table shows all four, including the 2 cents or more row where the sample falls to 6 games and the test says nothing at all. Reporting only the threshold that looked most interesting would have been the easiest way to mislead here.
- The two prices are built differently. Kalshi's number is the midpoint of the yes bid and ask on its last hourly candle before first pitch. Polymarket's is that venue's own price series at hourly fidelity. Part of any systematic gap between them is construction rather than forecast.
One window of one sport is not a finding about prediction markets in general. It is a record of what these two venues did over 30 days of baseball, published with the count attached so it can be checked and so the next window can be compared against it.
The sportsbook column, and where it is blank
The same comparison against a sportsbook price is only honest where we hold a real book capture taken before the game started. We do, on 197 of the 371 games, from a mean of 11 books quoting at once, devigged and taken as a median across those books.
On the other 174 games that column is blank. It is left blank. Our historical odds archive stores a daily fixed-hour grid whose timestamps are not real capture moments, so filling the gap from it would produce a number that looks like a pre-game book price and is not one.
Where all three sources were captured within 60 minutes of each other, which is 96 games, the Brier scores were 0.2448 for Kalshi, 0.2449 for Polymarket and 0.2446 for the book median. The three sit within 0.0003 of each other. The mean distance from the book median to Kalshi was 0.6 cents.
Across all 197 games with a book median the book's Brier score is 0.2388, which is lower than either venue's. That comparison is not fair and we are not making it: those book captures sit a mean 112 minutes before first pitch while the venue prices sit 24 minutes before, and they cover a different subset of games. The 96 game like-for-like subset above is the only three-way comparison on this page that compares the same games at the same distance from first pitch.
What this measurement cannot see
The grid is what it is, and the page is worth less than nothing if it hides that.
- These are not closing prices and we do not describe them as such. Both venue histories arrive at hourly resolution, so the last observation before first pitch sits a median 15 minutes out on Kalshi and 15 minutes out on Polymarket. Whatever the two venues did in the final minutes before a game, this page did not see it and makes no claim about it.
- The two venue prices are near simultaneous; the book price is not. The median gap between the Kalshi and Polymarket observations on the same game is 0.17 minutes, so calling their difference a disagreement at a moment is fair. The book capture is a mean 112 minutes before first pitch, so the three-way comparison is restricted to the aligned subset.
- It is one side of one market. Every probability here is the away team's chance of winning, on the moneyline only. Polymarket's history holds one side per game, and that side is the away team, so the home number would be an inferred complement rather than something either venue was observed to show. Totals and spreads are out of scope.
- It is 30 days. The window starts where the two venues' histories overlap and ends where our capture of them ends. It covers 371 of the 455 MLB games played in that span; the rest are games one venue did not list or did not price before first pitch.
- 5 games are recorded and deliberately not graded. They are doubleheaders where the two halves went to different teams and our results table stores no start time, so there is no honest way to say which half the priced game was. They stay in the table marked ungraded rather than being quietly dropped.
- 15 more games are captured and not yet graded. Their result has not reached our results table yet. They are counted here rather than in the figures above, and they will be graded when the result lands.
How this was measured
- Window: 6 July 2026 to 12 August 2026, 30 days of MLB games. n: 371 graded games.
- Price: each venue's last observation strictly before first pitch. Kalshi is the midpoint of the yes bid and ask on the last hourly candle; Polymarket is its own price history at hourly fidelity. Both refer to the away team winning.
- Book median: the median devigged away price across the books quoting at the last real pre-game board capture, minimum three books, mean 11. Two-way vig is removed proportionally. Rows without such a capture are stored as null and reported as null, never imputed.
- Comparability gate: the two venue observations must be within 60 minutes of each other and each within 180 minutes of first pitch. 9 of 398 built rows fail that gate and are excluded from every figure above.
- Grading: the realised result comes from our own results table. Grading fills empty fields once and never rewrites them; the database refuses an update to a graded row.
- Provenance: figures produced by
seo/f70_divergence_facts.pyover a session Postgres refuses to write through, and stamped into this page byseo/build_f70_divergence_page.py.
Next steps
Frequently asked questions
Do Kalshi and Polymarket price MLB games the same way?
In this window they mostly landed on the same number. Across 371 graded games the two venues showed an identical pre-game price on 212 of them, and the largest difference on any single game was 3.0 cents.
Which prediction market is more accurate on baseball?
On this sample neither is measurably more accurate. The Brier score was 0.2415 for Kalshi and 0.2421 for Polymarket across 371 graded games, a gap far smaller than the noise at this sample size. The counts for the subset where they disagreed are in the table above, with the sample size on every row.
Are these closing prices?
No. Both venue histories arrive hourly, so the last price before first pitch sits a median 15 minutes out. This page makes no claim about what either venue did in the final minutes before a game, because our data cannot see it.
How do the prediction markets compare to sportsbooks?
On the 96 games where a book capture sits within 60 minutes of the venue prices, the Brier scores were 0.2448, 0.2449 and 0.2446 for Kalshi, Polymarket and the book median. On the games where we hold no real pre-game book capture the column is left blank rather than estimated.
Related: MLB prices by book · How often lines move · Track record
Every figure on this page is a measurement of prices that already existed, on games that have already finished, produced by a committed and re-runnable script over a read-only copy of our archive. It describes what two venues did in one window. It is not a forecast and not betting advice: Nebula Insights publishes no picks, no selections and no recommended wagers. 18+ only.