Where prediction markets are in 2026
An honest snapshot of the prediction market industry after the 2024 cycle, Kalshi's CFTC win, and the slow arrival of cryptographic settlement primitives.
The prediction market industry looks different in mid-2026 than it did at the start of the 2024 US election cycle. Three structural changes have settled in, and another three are visibly in motion. This is the honest snapshot — what shifted, what did not, and where the architectural opportunity sits for the next round of builders.
What 2024 actually demonstrated
The 2024 US election cycle was the largest prediction market event in history by volume. Polymarket alone settled billions in notional across political markets. The cycle proved several things that had been speculative:
Prediction markets work at scale. The price discovery on tier-one political contracts was sharp, fast, and survived genuinely high engagement. Polls were misleading; markets were less misleading. The information value held up.
Mainstream attention is real. Polymarket markets were cited in mainstream financial press, in academic economics papers, and in election-night broadcasts. The category exited the crypto-native niche it occupied for a decade.
Settlement is the bottleneck. Several high-profile markets generated public disputes about whether trigger conditions were met. UMA optimistic oracle handled some cleanly; Polymarket Inc.’s editorial discretion was invoked on others. The disputes did not undermine confidence in the markets as a category, but they showed clearly that the settlement layer is the system’s weakest point.
The third lesson is the one that drives the next architectural cycle. If settlement is the bottleneck, the next protocol to win the long tail of markets will be the one with the best settlement layer. That is the bet behind the zkTLS-based wave Mentat is part of.
What Kalshi’s CFTC arc demonstrated
In parallel with Polymarket’s growth, Kalshi spent the 2024-2026 period navigating CFTC regulation. The end state — Kalshi operating as a CFTC-regulated designated contract market with a meaningful slate of political event contracts — was a meaningful win for the regulated path. Kalshi’s user base grew substantially during this period.
What Kalshi demonstrated is not “centralized regulated markets are better.” It is that the regulatory perimeter is real and there is genuine institutional demand for prediction markets inside that perimeter. Kalshi’s model — staff-listed, KYC-mandatory, centrally-settled — is the correct architecture for that demand.
It is not the architecture for the long tail. Permissionless creation, global access, and cryptographic settlement are the architecture for the long tail. Both can coexist; one is not displacing the other. Mentat is built for the second category; Kalshi serves the first.
Where the long tail is going
The interesting structural shift is not on the tier-one markets (US elections, Fed rate decisions, top-five sports leagues). Those are well-served today, with various tradeoffs between Polymarket and Kalshi.
The interesting shift is on the long tail. There are categories of useful markets that no centralized operator will ever list because the addressable user base per market is too small to justify staff time. A market on a specific corporate earnings call’s revenue beat. A market on a specific climate metric for a specific region. A market on a specific protocol launch milestone. A market on a specific scientific paper’s conclusion being replicated within two years.
These markets have value to small communities of participants. They have negligible value to a centralized operator’s staff queue. They will only exist if creation is permissionless and authoring quality is maintained without staff bottlenecks.
The AI authoring pipeline is the bet on the long tail. If creating a high-quality market spec drops from “two hours of expert work” to “fifteen seconds of AI work plus five minutes of curator review,” the addressable set of viable markets expands by orders of magnitude. That is not a marginal product improvement — it is a category shift.
What is shipping in 2026
A non-exhaustive list of what the broader prediction market space is shipping this year:
zkTLS infrastructure is maturing. TLSNotary, Pluto, and several other teams are shipping production-ready prover stacks. The proof generation cost has dropped substantially over 2025. On-chain verification is genuinely feasible for a broader set of trigger circuits.
AI authoring is becoming standard. Multiple protocols beyond Mentat are building AI-assisted market creation. The pattern is converging because the unit economics force it: human authoring at scale is uneconomic; AI authoring with human gates is.
Solana fee economics are reshaping what is viable. The shift of new prediction market builds from EVM to Solana is not branding — it is fee floors. A market with $200 of liquidity and 50 trades cannot pay $5 in EVM gas; it pays sub-cents on Solana. Long-tail markets fundamentally require sub-cent settlement.
Account abstraction is making wallets less of a barrier. Email-based wallet abstractions (Privy, Dynamic, custom Solana implementations) are removing the “install MetaMask first” onboarding friction that capped the addressable user base of crypto prediction markets to crypto-native users. The next million prediction market users will not have heard of a seed phrase.
Cross-protocol composability is showing up. Prediction market outcomes as SPL tokens (or ERC-20 equivalents) plug into lending markets, structured products, and on-chain derivatives. Augur tried this; the primitives are now mature enough that it works.
What is still broken
A few honest unsolved problems:
Dispute resolution at scale. Even with zkTLS handling the common case, edge-case disputes still need a process. UMA’s DVM design is one answer; multisig juries are another; off-chain arbitration is a third. None of them are obviously correct. Mentat is shipping a configurable hook for M4-M5 and letting deployments pick.
Curator economics at scale. A protocol that depends on human curators needs to pay them. The funding models we describe in the curation post — settlement fee share, creator stake forfeit, protocol grants — are credible but unproven at scale. The first prediction market protocol to run a thousand active markets with sustainable curator economics will have learned something important.
Liquidity bootstrapping for new markets. The long tail’s economics work only if liquidity flows to new markets quickly. Polymarket spent a half-decade and substantial venture capital building this on Polygon; Kalshi built it through institutional market makers under CFTC umbrella. New protocols start cold. Solving liquidity bootstrapping for permissionless markets is genuinely hard.
KYC, regulation, and the global perimeter. Every protocol has to decide how much regulatory perimeter to take on. Permissionless protocols generally punt this to deployments; deployments end up navigating jurisdiction-by-jurisdiction. The state of play is not stable and reasonable people disagree about where it lands.
What this means for builders
If you are building in this space in 2026, a few observations worth taking seriously:
The architecture choice is durable. Whatever settlement model you pick, you live with for years. Picking centralized arbitration in 2026 is picking an architecture that the next cycle will look at the way the 2017 cycle looks at the “trust me, I’m a multisig” pattern.
The AI layer is becoming undifferentiated. Everyone is shipping AI-assisted authoring. The differentiation will be in how well the curator layer is designed, how well the schema enforces verifiability, and whether the protocol takes the human gate seriously. Shipping the AI is easy; shipping the system is hard.
Solana is winning the new builds. This is observable. The fee economics matter and the developer ecosystem has matured. There are exceptions (some EVM-native protocols extend their existing footprints), but new prediction market protocols are mostly Solana-first.
Open source is winning the new builds. The proprietary protocols (Polymarket, Kalshi) have moats from their categories and operational footprints. New protocols entering the space are choosing open source because the ecosystem expects it and because integration partners will not commit to closed-source infrastructure.
The category is bigger than crypto Twitter thinks. Prediction markets are a primitive — like AMMs or NFTs — that will compose into many products, most of which will not be branded “prediction markets.” Information markets inside enterprise risk management. Forecasting markets inside R&D allocation. Compliance markets inside regulated financial products. The protocol layer for these is what we are all building toward; the user-facing products will be heterogeneous.
Where Mentat sits
We are one of several teams building toward the zkTLS-settled, AI-authored, permissionlessly-created future. We are not the only ones noticing the architectural opportunity. The competitive landscape includes several other promising protocols at various stages of build.
What we think Mentat does specifically well: the canonical schema, the curator console design, the explicit AI safety story, the Solana-native economics, the commitment to open source. What we think we still have to prove: liquidity bootstrapping post-mainnet, the proof verifier circuit performance at scale, the curator economic model in production.
The honest version is: it is the early innings. The 2024 cycle established that prediction markets are a category. The 2026-2028 cycle will establish which architectures win for which categories. We are betting that cryptographic settlement plus AI-assisted authoring plus permanent human curation is the right combination for the long tail.
We could be wrong. But the case is clear enough that it is worth building, and the alternative architectures have not stayed cleanly ahead. That is what makes this an interesting moment to be in the space.