Mentat vs Azuro: liquidity primitives vs full-stack market protocols

A frequent question we get is whether Mentat competes with Azuro. The short answer: not directly, but with overlap worth being honest about. The longer answer is a useful lens on how prediction market infrastructure has split into layers.

What Azuro actually is

Azuro is a liquidity protocol for on-chain prediction markets, primarily sports. The protocol provides pooled liquidity to partner front-ends (DivvyBet, BetSwirl, others) that handle UX, customer acquisition, and category specialization. Azuro’s economic primitives are:

• Liquidity Pool (LP). Capital providers deposit, earn fees from settled markets.
• Condition. A market definition with outcomes and odds.
• Outcome. Settlement entry pointing to a data provider that resolves the condition.
• Core. The settlement contract that pays out based on the configured data provider.

Data providers for outcomes are configured per partner. Trust in the resolution path is delegated to the chosen provider — typically a sports data API operator, sometimes an oracle network, sometimes the partner front-end itself.

Azuro is well-architected for its problem space. Sports betting is a category where outcomes are reported promptly by professional data providers (Sportradar, Stats Perform, Genius Sports), where odds compilation is a specialist skill, and where liquidity needs to be pooled across many simultaneous events. The protocol’s separation of concerns — liquidity layer vs front-end vs data provider — fits that reality.

What Mentat is, in contrast

Mentat is a full-stack prediction market protocol: schema, authoring, curation, on-chain programs, settlement layer. The protocol cares about how markets are created (AI agents + curators), how they are specified (canonical schema with explicit triggers and source allowlists), and how they are resolved (zkTLS proofs verified on-chain).

The settlement model is the key difference. Azuro markets resolve when the configured data provider reports an outcome. Mentat markets resolve when a zkTLS proof verifies on-chain against a source declared in the market spec. The chain trusts cryptography, not the configured provider.

These are not the same product:

• Azuro is infrastructure for operators. It assumes a sophisticated front-end takes care of authoring, marketing, and category expertise.
• Mentat is infrastructure for markets themselves. It includes the authoring and curation layer because permissionless market creation is the design goal.

Where the comparison sharpens

Despite the layer difference, there is real overlap on two dimensions worth examining.

Resolution model

Both protocols need a way to bridge real-world outcomes to on-chain settlement. Azuro delegates to a data provider. Mentat verifies a cryptographic proof. The tradeoffs:

Azuro (delegated provider). Fast, simple, requires trust in the provider. Excellent fit for sports because the data providers are professional operators that the market makers and bettors already implicitly trust. Worse fit for outcomes where no single specialist provider exists.

Mentat (zkTLS). Slower to settle (proof generation latency, on-chain verification), more complex to operate, requires no trust in any single delegated party. Excellent fit for outcomes that have a canonical public source (APIs, official statements, government data feeds). Worse fit for sports specifically because the canonical sources for sports are typically the licensed feeds (Sportradar, Stats Perform) that charge commercial rates and may not be cleanly proveable via zkTLS without commercial agreements.

This is not a coincidence. The two protocols optimize for different category profiles.

Permissionless market creation

Azuro markets are configured by partners. Most partner front-ends offer a curated set of events — the front-end’s commercial proposition is its category specialization and event selection. Permissionless creation at the partner level is rare.

Mentat is designed for permissionless creation. The AI pipeline plus curator console is the protocol’s answer to “anyone can deploy a market without losing quality.” A hypothetical Mentat-style overlay on Azuro is not coherent because Azuro liquidity is per-pool and pooled liquidity does not handle the long tail of arbitrary markets economically.

A hybrid scenario worth considering

There is a coherent architectural scenario where the two protocols complement each other rather than compete.

Imagine a multi-chain Mentat deployment that supports EVM in addition to Solana. The Solana deployment uses native on-chain liquidity per market. The EVM deployment could, in principle, use Azuro pools as the liquidity layer while keeping Mentat’s authoring + curation + zkTLS settlement on top.

In that scenario, Azuro provides the LP primitive and Mentat provides the protocol primitive. The data provider on the Azuro side becomes “Mentat’s zkTLS proof artifact” — which is admittedly an unusual configuration but architecturally consistent.

We are not building this today. M3 is Solana-native and M4 is the proof service. EVM is a downstream consideration. But the architectural shape is interesting because it suggests the two protocols can plausibly coexist in a stack rather than fight for the same slot.

What Azuro is genuinely better at

Sports specifically. The category demands fast settlement, professional data feeds, deep per-event liquidity, and specialized odds compilation. Azuro’s architecture is purpose-built for it and the partner ecosystem (DivvyBet, BetSwirl, others) is delivering polished product. Mentat is not optimized for sports and we should not pretend we are.

Live betting (in-play). Sports betting has a large live segment where odds shift in real time during an event. Azuro and its partners handle this via fast oracles and rapid settlement. Mentat’s zkTLS settlement model is not designed for live in-play; the proof generation timing alone makes it impractical for sub-second resolution windows. Different category, different architecture.

Existing partner network. Azuro has working partner front-ends and active LP capital. Mentat is pre-mainnet.

What Mentat is genuinely better at

Categorical event markets with canonical public sources. “Will the Federal Reserve raise rates by 25bps on this date” has a canonical source (FOMC press release). zkTLS proves it. No data provider needed.

Long-tail markets where no specialist data provider exists. The Mentat authoring pipeline can scaffold a market against any public source the proof service can attest to. The economic model does not require a partner front-end to commercially negotiate a data provider relationship.

Markets with adversarial dispute surface. UMA’s experience and Azuro’s sports model both struggle when the trigger condition is interpretable. Mentat pushes interpretation upstream into the authoring layer (Validator agent + curator) and resolves only triggers that are deterministic. The dispute surface is much smaller.

Open source self-hosting. Azuro contracts are open source but the protocol stack is not architected for full self-host. Mentat is — backend, agents, programs, indexer, proof service all Apache-2.0 and operable independently.

How to think about choosing

If you are operating a sports betting product, choose Azuro. The architecture and the partner network are both correct for the category.

If you are building a prediction market product for categorical events — politics, economics, technology milestones, science, governance — choose Mentat. The authoring pipeline and zkTLS settlement are both correct for those categories.

If you are building an infrastructure protocol that wants to support multiple verticals, the two stacks can compose. We do not see Azuro and Mentat as zero-sum.

The most useful framing is this: prediction market infrastructure has bifurcated into a liquidity layer (Azuro and similar) and a protocol layer (Mentat and similar). Both layers are necessary for the space to grow beyond Polymarket’s monoculture. The interesting product question is not “which one wins” but “which stack does my product require and what does the composition look like.”