12 Cross-Chain Swap Venues Compared
Rift Research
Last updated
There is no single best cross-chain swap venue. The strongest quote depends on the asset, trade size, available liquidity and total fees. Intent networks, native AMMs and orderbooks can each serve different routes. Rift compares venues with no added router fee and enforces noncustodial execution. This page compares twelve venues on the five dimensions that determine which one to use.
How this comparison is scored
Every venue is scored on the same five dimensions: trust model, asset model, cost structure, settlement time and failure mode.
Mechanism descriptions come from each protocol's own documentation, linked at the foot of the page.
Two separate things are called Rift, and only one is in this table.
Rift OTC is an execution venue. It competes for routes alongside every other venue here and is scored on the same basis, including on cost. It charges 0.1% taker, and only when it wins the route on rate after that fee is applied.
The Rift router is not a venue and is not in this table. It charges nothing on any route. It appears in the fee model section below because a router's fee structure changes what you pay regardless of which venue fills.
Where these venues beat Rift states where Rift is not the right answer, and commits to expanding as data lands.
How do the 12 venues compare?
| Venue | Type | Trust model | Asset model | Cost structure | Typical settlement | Primary failure mode |
|---|---|---|---|---|---|---|
| Across | Intent + solver | UMA optimistic oracle plus bonded relayers. Fraudulent fills are disputable and the bond slashable during the challenge window | Canonical assets only. No wrapped or synthetic tokens | Relayer spread plus LP fee. No separate protocol take on top of those two components | 2–15 seconds on major EVM routes | Oracle dispute failure. Relayer inventory exhaustion on thin routes |
| Relay | Intent + solver | A relayer fronts liquidity and settles against the source chain. Single relayer today. A bonded multi-relayer protocol is stated as forthcoming | Native assets on destination | Solver spread, quoted upfront | Seconds on small transfers | Single-relayer concentration until the bonded network ships |
| NEAR Intents | Intent + solver | Atomic through the intents.near verifier contract. Solvers bring their own capital | Native assets | Solver spread. No LP overhead | Seconds | Solver competition thins on unusual pairs. NEAR chain dependency |
| Mayan (Swift) | Intent + solver | Layered on Wormhole for messaging. Inherits Wormhole's guardian trust model | Native assets on destination | Solver spread | Under 12 seconds on EVM-to-Solana | Inherits the messaging layer's failure modes |
| deBridge | Intent + solver (DLN) | Solver network with its own messaging layer | Native assets | Solver spread | Seconds on major routes | Solver liquidity on long-tail routes |
| Garden | HTLC atomic swap + solver | Hashed timelock contracts. Both legs complete or the asset refunds. Solvers are independent operators | Native assets. No wrapped tokens | Solver spread | 5–10 minutes on Bitcoin routes | Solver-layer compromise. Two incidents on record. Contracts held. No user funds reported lost1 |
| THORChain | Native AMM | An independent L1. Vaults are controlled by a bonded node set signing outbound transfers via threshold signatures. Security is economic: bonded value must exceed value at risk | Native assets, no wrapping. RUNE is the settlement asset | Slip-based liquidity fee plus outbound network fee. Documented range 5–50 bps | Minutes. Protocol documentation cites roughly 2,640 seconds end to end in its BTC worked example | Economic security failure if bonded value falls below value at risk. Slip on thin pools |
| Chainflip | JIT AMM | 150 validators. A 100-of-150 threshold signature supermajority must sign. No single validator holds vault keys | Native assets. USDC as base pair | JIT market-maker spread, priced per swap | Minutes on Bitcoin routes | Validator set compromise. Depth constrained by protocol TVL |
| CCTP (Circle) | Native burn and mint | Circle's attestation service. Centralised issuer, no third-party validator set | Native USDC only. No wrapped representation | No protocol fee on standard transfers. V2 Fast Transfer carries a fixed finality fee reported at 1 bp | 15–20 minutes standard. Faster on V2 Fast Transfer | Issuer dependency. USDC only, so no route for other assets |
| Unit | Lock and mint into Hyperliquid | Guardian network using MPC threshold signatures, 2-of-3 for critical operations. Guardians are not publicly named. No public audit confirmed | 1:1 representation on Hyperliquid | No protocol fee. Two network fees, typically around $1–2 combined | Roughly 25 minutes for Bitcoin, gated on 2 confirmations | Guardian collusion or compromise. A guardian outage in April 2025 delayed Bitcoin withdrawals. Deposits below 0.0003 BTC are unrecoverable (source) |
| Hyperliquid | Central limit orderbook | Onchain orderbook. Assets must be bridged in first, typically via Unit | Spot balances on Hyperliquid | Taker and maker fee schedule, plus a one-time activation fee on first USDC receipt | Seconds once assets are on the venue | Bridge-in dependency. You inherit Unit's trust model to reach it |
| Rift OTC | Bilateral OTC, competing as a venue | Settlement runs in Intel TDX enclaves on GCP rather than on a chain. Upgrades gated by a hardware-key multisig and a 7-day timelock. Onchain reads verified 2-of-3 across independent RPCs inside the enclave. Trade-off: trust in the hardware vendor and data centre operator, in exchange for no staking collateral. Not permissionless | Native assets on both sides | 0.1% taker, charged only when Rift OTC wins the route on rate after that fee is applied | Counterparty quote, then settlement on the slower leg | Counterparty non-performance within the quote window. TEE or data centre compromise. Route coverage limited to live pairs |
How does a router's fee model change what you pay?
A venue's fee is not the only thing you pay. If you reach that venue through a router, the router's own fee model applies on top.
Most routers add a markup on every trade. A spread is applied to the venue's rate before you are quoted. You see the marked-up number, never the underlying one, so there is nothing to compare against. Reported industry range is 0.05% to 0.3%, applied whether or not the routing improved your outcome.
The Rift router charges nothing. No fee, no spread, no markup, on any route. You receive the winning venue's rate, less the gas of the additional hop that routing necessarily adds.
Rift OTC is a separate venue, not the router. When Rift OTC wins a route it charges 0.1% taker, and it only wins when it is cheapest after that fee. The router earns nothing from routing to it.
How to check any router, including this one. Quote the same route on the router and directly on the venue it says it is using. The difference, less the gas of the extra hop, is the markup. On a zero-fee router the two should match.
How do you choose between them?
Match the venue to the constraint that binds you. Most people optimise for headline fee. The binding constraint is usually asset support or size.
| If you are moving | Use | Because |
|---|---|---|
| USDC between EVM chains, under $50k | An intent rail. Across, Relay, deBridge | Relayer competition is fierce and fixed gas is small relative to the transfer |
| USDC above $50k | CCTP | No protocol fee on standard transfers, no liquidity cap |
| Native Bitcoin | THORChain, Chainflip, Garden | All settle native. No wrapped representation |
| Into or out of Hyperliquid | Unit, then the orderbook | You accept a 2-of-3 guardian assumption to reach orderbook depth |
| Assets outside USDC and EVM | NEAR Intents | Solvers source their own liquidity, covering pairs the AMMs do not |
| Above eight figures | Compare full-size routed, orderbook and OTC quotes | Depth and negotiated pricing can help, but actual fees and settlement costs decide the net result. See How Do You Swap $10M of BTC Cross-Chain? |
| You do not know which of the above wins today | A router | The winning venue changes with depth, gas and inventory. A router quotes across all of them and executes the best one. Check what the router charges: most add a markup on every trade, the Rift router adds nothing |
When a router is not worth it. If you already know which venue wins your route and you run it repeatedly, go direct. You save the gas of the extra hop. Routing earns its cost when the answer changes between trades, which on most routes it does.
What are the five trust models, and what does each one assume?
Every venue falls into one of five categories. The category determines what has to go wrong for you to lose money.
Economic security. THORChain. A bonded node set controls the vaults and signs outbound transfers. The assumption is that bonded value exceeds value at risk, making an attack unprofitable rather than impossible. The security budget is observable onchain.
Threshold signature committees. Chainflip, Unit, and Mayan by inheritance through Wormhole. A committee holds key shares and a supermajority must sign. Committee size and identity are the questions that matter. Chainflip requires 100 of 150. Unit requires 2 of 3, and its guardians are not publicly named.
Optimistic verification. Across. Fills execute immediately and are verified afterwards. A dispute window allows a fraudulent assertion to be challenged and the proposer's bond slashed. The assumption is that at least one honest party watches.
Hardware enforcement. Rift. Code runs in an enclave rather than on a chain, with upgrades gated by a hardware-key multisig and a timelock. No staking collateral is required. The assumption is that the hardware vendor and the data centre operator behave. Not permissionless.
Atomic settlement. Garden through HTLCs. NEAR Intents through its verifier contract. Either both legs complete or the transaction reverts. The strongest contract-level guarantee here.
The rule that matters: contract-level guarantees do not extend to the off-chain coordination layer that intent and solver systems depend on. Garden's two incidents both originated there. The contracts held in both.
What does each venue actually cost?
Cost structure is in the comparison table above: how each venue charges you.
Measured figures are not on this page. Full six-component decomposition at $10k, $100k and $10M, measured in a single window, is in What Does a Cross-Chain Swap Actually Cost?.
The short version: the protocol fee is only one component of what you pay. Compare final output after venue fees, spreads, price impact, network costs and any router markup.
What is excluded from this comparison, and why?
- Wrapped-asset bridges. A wrapped asset carries counterparty risk for as long as you hold it, not just for the trade. Different risk category, not comparable on cost.
- Message-passing layers. LayerZero, Wormhole and Hyperlane are transport primitives, not venues. Several venues here run on top of them, and the rail's trust model passes through.
- DEX aggregators. Velora, formerly ParaSwap, and similar source liquidity across protocols rather than providing a venue.
- Centralised exchanges. These are custodial trading platforms rather than the onchain execution venues compared here. They belong in a cost comparison, but a tight orderbook price does not establish the best net rate. Rift enforces noncustodial execution and can often offer better output after retail CEX trading and withdrawal fees. The lowest institutional fee tiers and negotiated OTC pricing for very large size can change the result. Use your actual account fees: Coinbase and Kraken both publish guidance on tiered pricing.
Where these venues beat Rift
Rift OTC is not the correct answer on several routes, and neither is routing through Rift.
- USDC between EVM chains, at any size. CCTP burns and mints natively against Circle with no protocol fee on standard transfers. Nothing routing on top of it beats it on that pair.
- Chains Rift does not support. Rift supports Bitcoin, Ethereum, Arbitrum and Base. Solana, TON, Tron, Sui and Stellar require NEAR Intents, deBridge or Symbiosis. Relay covers 85+ networks.
- Transfers under $1,000 on major EVM routes. Fixed costs dominate. A direct intent rail has fewer steps and fewer fixed costs.
- Any route where another venue wins. The router takes nothing, so going direct is cheaper by exactly the gas of the extra hop.
- Anything needing a wrapped representation as the end state. Rift settles native. If wBTC is the intended holding, a wrapping bridge is the direct path.
Which venue is fastest?
On any route with a Bitcoin leg, the source chain governs settlement time, not the venue.
- Under 15 seconds. Across on major EVM routes. Mayan Swift on EVM-to-Solana. Relay on small transfers.
- Minutes. CCTP standard transfers. Chainflip.
- 5 to 30 minutes. Any Bitcoin leg. Unit waits two confirmations, roughly 25 minutes. Garden quotes 5–10 minutes. THORChain's documentation gives roughly 2,640 seconds end to end in its worked example.
- Route-dependent. Routing through Rift settles at the speed of whichever venue wins, plus the additional hop. It cannot be faster than the underlying venue.
Speed claims that exclude the Bitcoin leg are measuring the wrong thing.
Sources
Every mechanism and trust-model claim above is drawn from primary documentation.
- Across architecture and optimistic oracle settlement — UMA: https://blog.uma.xyz/articles/case-study-how-uma-secures-across-protocol
- Across hub-and-spoke topology and fill statistics — Eco: https://eco.com/support/en/articles/14799839-across-protocol-solver-architecture-deep-dive
- Relay settlement architecture — Relay docs: https://docs.relay.link/how-it-works/the-relay-solver
- NEAR Intents express/solve/settle model — LeoDex: https://leodex.io/learn/near-intents/what-is-near-intents
- Mayan Swift settlement times and Wormhole dependency — Eco: https://eco.com/support/en/articles/13017848-best-solana-bridge-for-2026
- Garden HTLC and intent flow — Garden docs: https://docs.garden.finance/home/fundamentals/how-it-works/intent-flow
- Garden July 2026 solver incident — Crypto.news: https://crypto.news/garden-finance-takes-app-offline-after-independent-solver-database-compromise/
- THORChain fee structure and streaming swaps — THORChain docs: https://docs.thorchain.org/technical-documentation/technical-deep-dive/fees
- THORChain worked swap example and timings — THORChain dev docs: https://dev.thorchain.org/swap-guide/quickstart-guide.html
- Chainflip validator set and JIT AMM — Chainflip: https://chainflip.io/blog/what-is-chainflip-complete-guide
- CCTP V2 fee mechanics — Circle CCTP V2 white paper: https://6778953.fs1.hubspotusercontent-na1.net/hubfs/6778953/PDFs/Whitepapers/CCTPV2_White_Paper.pdf
- Unit Guardian network and MPC TSS — Unit docs: https://docs.hyperunit.xyz/architecture/quickstart
- Unit deposit mechanics, confirmations and minimum deposit — Unit docs: https://docs.hyperunit.xyz/how-to/deposit
- Bridging fee component breakdown — Eco: https://eco.com/support/en/articles/11751649-crypto-bridging-fees-why-they-vary-10x
Footnotes
-
Garden's two solver incidents: approximately $11.4M in October 2025, and approximately $450,000 in July 2026 across Ethereum, Base, Arbitrum and BNB Chain. Both were attributed to compromised solver infrastructure rather than the HTLC contracts themselves. Garden reported no user funds lost in either case. ↩