2026 Senate Election Odds: The Races That Decide Control
The Senate is the chamber the 2026 midterms will actually be fought over. Kalshi's control market prices it near a coin flip, more than 6.2 million contracts traded, and the individual race markets underneath it are some of the biggest political markets on the exchange, led by Texas at nearly 7 million contracts traded. We gave eight AI models the verified midterm registry, the full race-ratings table, live polling with per-pollster track records, and the seat math, no prices anywhere, and asked each for one coherent view: the probability of Democratic Senate control plus a Democratic win probability for each of the five priced races. Then each model read the others' anonymized reasoning and revised. The result is one of the sharpest market disagreements we have published, and it is a disagreement about arithmetic, not mood.
This is a neutral, data-driven analysis of market prices, ratings, polling, and historical patterns. It does not advocate for any party, candidate, or outcome.
The Quick Answer
Kalshi prices Democratic Senate control at 46.5 cents, nearly a coin flip. Our eight-model, price-blind panel blends to just 16.7% after a revision round, because the map forces a near-parlay: Democrats need +4 net with only four Republican-held seats rated competitive, while defending three of their own. Race by race, the panel and the market are closest on Texas (38.1% vs 46.5 cents) and furthest apart on North Carolina (68.6% vs 90.75 cents), Georgia (66.0% vs 90.35 cents), and Iowa (16.1% vs 42.5 cents). The full race board, every seat's number, and both honest explanations for the 30-point control gap are below.
The Market At A Glance
| Venue | Kalshi, a CFTC-regulated event-contract exchange (18+; availability varies by state, as of August 2026) |
| The Control Market | "Which party will control the Senate?" (CONTROLS-2026), two mutually exclusive party contracts, ~6.26M contracts traded |
| The Race Markets | Texas (~6.97M contracts), Maine (~3.65M), North Carolina (~1.33M), Iowa (~0.99M), Georgia (~0.97M), each a two-candidate market |
| Race Settlement | Per the rules, verbatim (Texas Democratic contract): "If a representative of the Democratic party is sworn in as a Senator of Texas for the term beginning in 2027, then the market resolves to Yes" |
| Key Dates | Election day November 3, 2026; Georgia runoff December 1 if no majority; race markets expect expiration January 4, 2027; the control market February 1, 2027 |
| Prices Below As Of | August 1, 2026, 12:39 UTC (live bid/ask midpoints) |
These are model estimates, not predictions of fact and not financial advice. Kalshi event contracts trade on a CFTC-regulated exchange; you must be 18 or older and in an eligible state to participate.
The Control Picture In One Table
| Market | Kalshi (mid) | AI Blend | Gap |
|---|---|---|---|
| Democratic Senate control | 46.5¢ | 16.7% | −29.8 |
| Texas: Talarico (D) beats Paxton (R) | 46.5¢ | 38.1% | −8.4 |
| Maine: Jackson (D) beats Collins (R) | 62.5¢ | 49.3% | −13.2 |
| North Carolina: Cooper (D) beats Whatley (R) | 90.75¢ | 68.6% | −22.1 |
| Iowa: Turek (D) beats Hinson (R) | 42.5¢ | 16.1% | −26.4 |
| Georgia: Ossoff (D) holds vs Collins (R) | 90.35¢ | 66.0% | −24.4 |
These are model estimates, not predictions of fact and not financial advice. The blend is the equal-weight mean of eight models' final numbers after a revision round. The panel sits below the market on every line; the section on where that pattern comes from, including the data limitation that could explain part of it, is further down.
Every Seat's Number
| Seat | P(D control) | Texas | Maine | North Carolina | Iowa | Georgia |
|---|---|---|---|---|---|---|
| Claude Fable | 20.0 | 38.0 | 52.0 | 70.0 | 18.0 | 68.0 |
| Claude Opus | 17.0 | 38.0 | 51.0 | 80.0 | 15.0 | 76.0 |
| Claude Sonnet | 20.0 | 38.0 | 46.0 | 62.0 | 19.0 | 61.0 |
| ChatGPT | 19.0 | 38.0 | 51.0 | 67.0 | 19.0 | 66.0 |
| Gemini Pro | 16.0 | 38.0 | 50.0 | 68.0 | 15.0 | 65.0 |
| GLM | 15.0 | 38.0 | 48.0 | 62.0 | 20.0 | 60.0 |
| Kimi | 16.0 | 37.0 | 48.0 | 65.0 | 18.0 | 62.0 |
| DeepSeek | 11.0 | 40.0 | 48.0 | 75.0 | 5.0 | 70.0 |
These are model estimates, not predictions of fact and not financial advice. Numbers are each seat's final board after revision. Two convergences stand out: Texas, where all eight seats landed within three points of each other, and Maine, where two full rounds of analysis ended at 46-52 with reasons stated. That Maine cluster is not a default; it is the finding, and the Maine section below walks exactly why the edges net out to a coin flip.
The Map: Why +4 Is The Whole Story
The registry facts every model anchored on: 35 seats up in 2026 (33 regular plus 2 special elections), Republicans defending 22 and Democrats 13, with Democrats needing +4 net for a majority. The ratings table (Cook, July 1; Sabato, June 11) lists exactly four Republican-held seats as competitive: open North Carolina (Lean D flip), Maine (Tossup), Alaska (Tossup), and the Ohio special (Tossup). It simultaneously lists three Democratic defenses: open Michigan (Tossup), Georgia (Lean D), and open New Hampshire (Lean D per Sabato). The registry's own summary: the +4 path requires winning essentially all of the competitive Republican seats while losing none of those defenses; any single defensive loss forces a fifth pickup from seats currently rated safe or likely Republican.
Every seat walked that arithmetic explicitly, and the walks agree in shape. Claude Fable's version: "My board gives expected pickups of NC 0.72 + ME 0.55 + TX 0.40 + IA 0.22, plus unpriced AK ~0.42 and OH-special ~0.45, for ~2.76 expected flips; expected defensive losses are GA 0.30 + MI ~0.40 (tossup open) + NH ~0.25, roughly 0.95. That nets to about +1.8 seats — close to the historical -3.3 Senate midterm average for the president's party but short of the +4 threshold." — Claude Fable (from its round-one board; its final numbers moved slightly)
The environment side of the card points the other way and every model said so: generic ballot D+6.4, presidential approval net −13.4, and the 18-of-20 midterm base rate, with neither the 1998 nor the 2002 exception condition present. The 2018 precedent in the registry is what reconciles the two: Democrats flipped 41 House seats that night and still lost two Senate seats, because the Senate map, not the national mood, decides the chamber. That is the whole panel's model of 2026 in one sentence.
Texas: Paxton, Talarico And The Biggest Race Market On The Board
The biggest race market on the exchange at nearly 7 million contracts, and the one place the panel's data went deepest: the registry's Texas card carries the complete primary history, all nine post-primary head-to-head polls in the feed, ratings from eight raters, and the structural baseline.
The race exists in its current form because four-term incumbent John Cornyn was defeated in his own party's runoff on May 26, 2026, by Attorney General Ken Paxton, 63.8 to 36.2, a 27.6-point defeat of a sitting senator. Democratic state representative James Talarico won his primary outright with 52.4%. The card's fundraising line: Talarico announced $30 million in Q2, described as a record for a Senate candidate in the second quarter of an election year, against Paxton's $9 million. The nine head-to-heads since the runoff average Talarico +1.3 (+0.9 excluding partisan or aligned-sponsor polls), including a 51-48 Talarico result from Beacon Research/Shaw & Co. for Fox News in late July. Against that: the structural baseline. Texas voted R+13.68 for president in 2024, R+8.49 in the same year's Senate race, and no Democrat has won any statewide Texas office since 1994; the closest modern miss was Beto O'Rourke's 2.56-point loss to Ted Cruz in the strong Democratic midterm of 2018. The raters split cleanly: Cook, Sabato, and Inside Elections keep Texas at Lean or Likely R, while the aggregator models (The Economist, RealClearPolitics, Race to the WH) call it a Tossup.
The panel's read is the most unanimous number on the board: all eight seats between 37 and 40, blending to 38.1% against a 46.5-cent market. Gemini Pro's revision note frames the cut from its earlier 42: "While Talarico's polling and fundraising are stellar, the R+13.7 baseline and traditional raters' Lean R consensus suggest 42% in Texas was overly optimistic." — Gemini Pro. Read it this way: the panel accepts Texas is a real race, prices it above every rater's lean, and still refuses to call a state with a 30-year Democratic statewide drought a true toss-up in August.
North Carolina: The Lean-D Flip The Market Treats As Done
North Carolina is Democrats' best pickup chance by every source in the card: Thom Tillis is retiring, former two-term governor Roy Cooper is the Democratic nominee against former RNC chair Michael Whatley, and both Cook and Sabato rate it Lean D, the only Republican-held seat on the flip side of the line. The market takes that and prices Cooper at 90.75 cents.
That price is the panel's second-largest race disagreement: the blend is 68.6%, with even the most bullish seat (Claude Opus at 80) below the market. The panel's logic is about what "Lean D" historically means: a real favorite, not a formality, three months before an election in a state the registry shows is competitive at the presidential level. ChatGPT nudged its number up in revision, not down, "because both Cook and Sabato rate it a Lean D flip and it is the clearest R-held pickup," and still landed at 67. If the market is right about Cooper at 90, the control question tightens materially; if the panel is right, the single most important brick in the Democratic path is a two-in-three proposition, and the +4 parlay gets much harder. This race, more than any other, is where the 30-point control gap lives.
Maine: Collins, Jackson And An Honest Coin Flip
Five-term incumbent Susan Collins, Maine's longest-serving member of Congress and the Senate Appropriations chair since 2025, against former state Senate president Troy Jackson, in the one state on this board that votes by ranked-choice voting. Both raters call it a pure Tossup. The market does not: Jackson trades at 62.5 cents, and Kalshi's separate 16-candidate Maine person market prices identically (Jackson 62.5, Collins 37.5, everyone else at 0.15 cents or less), so the exchange treats the race as strictly two-horse.
The panel's two rounds ended with all eight seats between 46 and 52, a 49.3% blend, and the revision notes show the edges being weighed rather than skipped. Claude Opus trimmed from 55: "Collins is a five-term incumbent now chairing Appropriations against a lower-profile recruit, and both raters still call it a pure tossup despite the D+6.4 environment — so the environment tilt I applied above 50 was too generous." — Claude Opus. Claude Sonnet moved the other way, raising 42 to 46, because "the card states both Cook and Sabato rate it a true Tossup, not lean-R, which implies closer to even odds than I'd priced despite Collins' seniority." A D+6 national environment argues for Jackson; an incumbency brand nearly three decades old, now holding the Appropriations gavel, argues for Collins; the raters who watch this race professionally refuse to pick a side. The panel's coin flip is the analysis, and it makes the market's 62.5 cents on Jackson the quietly aggressive price on this board.
Iowa: The Open Seat The Raters Don't List
Joni Ernst is not seeking a third term, leaving an open race between congresswoman Ashley Hinson and Democratic state representative Josh Turek. The market prices Turek at 42.5 cents, remarkably close to its Texas number. The panel blends Iowa at 16.1%, its largest proportional fade on the board, for one registry-anchored reason: Iowa appears nowhere on the ratings card's competitive-race lists. Claude Fable's revision note names it directly: "Trimmed Iowa 22→18 because the card's competitive list omits Iowa entirely — a fact I had underweighted." — Claude Fable. DeepSeek went to 5%. An open seat in a D+6 year is not nothing, which is why most seats held 15-20 rather than DeepSeek's floor; but a 42.5-cent market price for a race no professional rater currently ranks competitive is the board's clearest case of market and ratings living in different worlds. One of them will be visibly wrong by November.
Georgia: Ossoff, A Runoff Rule, And 90 Cents
Jon Ossoff, the youngest sitting senator, defends against congressman Mike Collins in the one state here with a December 1 runoff if nobody clears a majority. Cook and Sabato rate it Lean D; the market prices Ossoff at 90.35 cents; the panel blends 66.0%. The shape of the disagreement is the same as North Carolina's: the panel pays Lean-D ratings like Lean-D ratings, roughly two-in-three, while the market pays them like Likely-D-plus. GLM's revision captures the panel's midpoint: it raised Georgia from 53 to 60 specifically to match "the Cook/Sabato rater consensus," and stopped there. Ossoff losing would not just cost Democrats a seat; per the map section above, a single defensive loss converts the +4 path into +5 from safe-rated territory, which is why the control blend is so sensitive to whether Georgia is a 66 or a 90.
Where The Panel Splits From The Market, And Both Honest Explanations
The pattern is one-directional: the panel is below the market on all six lines. Two explanations deserve equal billing.
The panel's case: parlay discipline. A 46.5-cent control price has to be built out of race probabilities, and the panel's argument is that the market's own race prices don't obviously support it without heroic assumptions about the unpriced races. The blend's expected Democratic pickups from the four priced Republican-held targets total about 1.7 seats before Alaska and the Ohio special, against live loss risk in Georgia, Michigan, and New Hampshire. The revision round even contained a real argument about how correlated these races are: Kimi raised its topline from 10 to 16 because "with race numbers like mine, moderate same-cycle correlation through the shared national environment roughly triples the sweep probability," while DeepSeek cut from 23 to 11 on the compound math running the other way. That spread, 11 to 20, is the honest uncertainty range for the parlay question, and all of it sits far below 46.5 cents.
The market's case: information the card did not carry. The panel's Texas read came from nine fetched head-to-head polls; its reads for Maine, North Carolina, Iowa, and Georgia came from ratings, candidate facts, and structure, because the polling feed carried no matched head-to-heads for those four races and the card said so explicitly. The market has no such constraint. If public and private state polling shows Cooper, Ossoff, or Turek running well ahead of their ratings, the market is pricing something the panel never saw, and the race-level gaps above would be the card's blind spot rather than the market's error. Prediction markets have earned respect on exactly this kind of question; our guide to whether prediction markets are accurate covers that record. Both stories will be graded on the same scoreboard.
How Race Prices Roll Up To The Control Price
Work One Example: From Five Races To One Number
Take the panel's blended race numbers and walk the map: NC at 68.6% and ME at 49.3% are the two live pickups among priced races, contributing about 1.2 expected flips; TX at 38.1% and IA at 16.1% add roughly 0.5 more; Alaska and the Ohio special, unpriced here but rated Tossup, plausibly add most of another seat. Call it 2.5 to 2.8 expected pickups. Subtract expected defensive losses in Georgia (34% by the blend), Michigan, and New Hampshire, and the expected net lands around +1.5 to +2 seats against a +4 requirement. Getting from "expected +2" to "actually +4" requires most tossups to break one way at once, which is why a topline probability can sit far below the average of its component races. That is the arithmetic engine of this entire page, and it is also why the chamber-control question and the race questions are different products: our House control board prices the other chamber's version, and the 2026 midterms balance of power board prices the House-Senate combinations as one partition.
The standard frictions apply before acting on anything here: Kalshi is a CFTC-regulated event-contract exchange, 18+ with availability varying by state; these markets hold capital into January-February 2027 because they settle on who is sworn in, not on election-night calls; and the two-sided books carry spread, so the mid is not the price you trade at. Every number in this piece lands in our backtest ledger and is graded against the real settlement on our public Model Verdict scoreboard. For the wider slate, the Kalshi picks hub tracks every board we have scored, politics markets for sports bettors covers the mechanics, our live odds screen and free expert picks today do the sportsbook-side work, and the same panel's longer-dated political boards are live for the 2028 Democratic nomination and the 2028 presidential election.
What Midterm Base Rates Say
The Senate-specific base rate is milder than the House's: the president's party has lost an average of 3.3 Senate seats across the 20 postwar midterms, and the chamber-divergence rows in the registry are the reason no model treated the national environment as decisive. In 2018, a blue-wave House year, Democrats lost two Senate seats while defending 26 of 35, ten of them in Trump-2016 states. In 2010, a minus-63 House catastrophe moved the Senate only six seats. In 2022, the post-Dobbs cycle, the president's party gained a Senate seat while losing nine in the House. Same nights, same voters, opposite chambers: the map is the machine. In 2026 the defense ratio is reversed from 2018 (Republicans defend 22 of 35), which is why this cycle is competitive at all, and the four-seat threshold is why competitive is not the same as even.
Primary-Night Re-Scores
The five races above have their nominees set. The remaining catalysts are scheduled: the last competitive-state nominees finalize through late summer, fall polling arrives in volume after Labor Day, and any rating change by Cook or Sabato moves the exact inputs this panel priced. Each of those triggers a re-score of this board on this same URL, with the changes logged, and the settlement retrospective lands here after the chamber is decided.
Want the re-scored board as the fall catalysts land? The email signup at the end of this page gets each re-run of this exact race-by-race board first, from the next ratings change through election week.
FAQ
When are the 2026 Senate elections and when do these markets settle? Election day is November 3, 2026, with 35 seats up (33 regular plus 2 special elections), Republicans defending 22 and Democrats 13. The race markets resolve on who is sworn in for the term beginning in 2027 (expected expiration January 4, 2027); the chamber-control market carries an expected expiration of February 1, 2027.
How does the Georgia runoff affect the Kalshi Senate market? Georgia schedules a runoff for December 1, 2026 if no candidate wins a majority on November 3. Because the Kalshi contract resolves on which party's candidate is sworn in for the 2027 term, a runoff delays resolution rather than changing what settles it.
The Bottom Line
The market says the 2026 Senate is a coin flip. The panel says it is a one-in-six, and the difference is not one number but five: a Texas race both sides agree is real, a North Carolina and a Georgia the market prices as nearly finished while the ratings say merely leaning, an Iowa the market prices as competitive while no rater lists it, and a Maine that two rounds of machine analysis ended exactly where Cook and Sabato have it, dead even. Watch the ratings, not the vibes: the single most price-moving event this board can experience before November is Cook or Sabato moving North Carolina, Georgia, or Maine, because those ratings are the load-bearing inputs under both the panel's numbers and, if the panel is right about the parlay math, under the market's coin flip too. Every number above is on the record and gets graded either way.
Hottest Prediction Markets Right Now
- 2028 Democratic presidential nominee · $166M traded
- Rocket Classic Winner · $85M traded
- 2027 Pro Football Champion · $52M traded
- Pro Baseball Champion · $41M traded
- Bitcoin price at the end of 2026 · $30M traded
- 2027 Pro Basketball Champion · $19M traded
Every market above links to our full AI model verdict; browse them all on the OddsShopper prediction markets hub.
Verdicts generated August 1, 2026, after a revision round in which each model read the others' anonymized reasoning. Prices as of August 1, 2026, 12:39 UTC. Panel: Claude Fable, Claude Opus, Claude Sonnet, ChatGPT, Gemini Pro, GLM, Kimi, and DeepSeek, all price-blind. This is a neutral, data-driven analysis and does not advocate for any party or outcome. These are model estimates, not predictions of fact and not financial advice.
