Decision guide // updated July 2026

OP Stack vs Arbitrum Orbit vs Polygon CDK vs ZK Stack: Choosing a Rollup Framework

A working comparison of OP Stack, Arbitrum Orbit, Polygon CDK, and ZK Stack in 2026, from a team that ships the app and infrastructure layer on top of them.

TL;DR

OP Stack is the most-adopted optimistic framework and a strong default for a general-purpose EVM L2, now at Stage 1 with permissionless fault proofs, with the Superchain for shared interoperability. Arbitrum's stack (formerly branded Orbit) gives you the most mature fraud-proof tech, but licensing it outside Arbitrum One carries a 10% revenue share. Polygon CDK is now a managed chain-as-a-service for ZK chains on the AggLayer, not a self-serve kit. ZKsync's ZK Stack builds sovereign ZK chains that join the Elastic Network. The framework is one decision; the data-availability layer, the sequencer, and who builds the surrounding infrastructure are the others.

Protofire builds the application and infrastructure layer on top of these stacks (explorer, oracles, indexing, Safe, first apps) and we are not affiliated with Optimism, Arbitrum, Polygon, or Matter Labs. Trust assumptions change fast here, so the stage and risk claims are anchored to L2Beat and the stacks' own docs, and each section names where a stack is the wrong choice. See our other decision guides.

Scorecard

At a glance

	01OP Stack	02Arbitrum (Orbit)	03Polygon CDK	04ZK Stack
Proof system	Optimistic (Stage 1)	Optimistic (Nitro, BOLD)	ZK (validity proofs)	ZK (validity proofs)
Withdrawal finality	~7-day challenge	~7-day challenge	Fast (ZK)	Fast (ZK)
EVM	Equivalent	Compatible	Compatible (zkEVM)	Compatible (custom compiler)
Interop model	Superchain (maturing)	Arbitrum ecosystem	AggLayer (maturing)	Elastic Network
Delivery	Open-source SDK	Open-source SDK	Managed (chain-as-a-service)	Open-source SDK
Cost model	No protocol fee	10% revenue share (AEP)	Work with Polygon Labs	No protocol fee
Native AA	Via EIP-7702 / 4337	Via EIP-7702 / 4337	Via EIP-7702 / 4337	Protocol-native

In detail

The four frameworks

OP Stack

The most-adopted optimistic stack

Strengths

+Largest optimistic ecosystem, with the Superchain for shared interoperability
+Now Stage 1 with permissionless fault proofs live on mainnet (Base, OP Mainnet)
+EVM-equivalent, so contracts behave as they do on Ethereum

Trade-offs

−Optimistic proofs mean a ~7-day challenge and withdrawal window
−Superchain native interop is still maturing, not yet fully live
−Defaults to a single sequencer, like the others

OP Stack is Optimism's open-source toolkit for EVM-equivalent rollups: it standardizes consensus, execution, and settlement, so you assemble a chain rather than write one, and it has the largest optimistic-rollup ecosystem. It uses fault proofs with a roughly seven-day challenge window, and OP Mainnet and Base are now Stage 1 with permissionless fault proofs (verifiable on L2Beat), a meaningful maturity milestone rather than the trust-me setup optimistic rollups once were.

Its distinguishing feature is the Superchain, a set of OP Stack chains (Base among them) that share a standard and are moving toward native interoperability, though that interop is still rolling out rather than fully shipped. Because OP Stack is EVM-equivalent, contracts and tooling behave as on Ethereum. For a general-purpose EVM L2 that wants the deepest tooling and the biggest ecosystem, it is a strong default.

Arbitrum (Orbit)

The most mature fraud-proof stack

Strengths

+Built on Arbitrum Nitro, the most-used optimistic technology, with BOLD permissionless fraud proofs
+High EVM compatibility and mature tooling
+Runs as a full rollup or lower-cost AnyTrust (DAC) chain, with a custom gas token

Trade-offs

−The AEP license takes 10% of protocol revenue for chains outside Arbitrum One and Nova
−Optimistic challenge window, like OP Stack
−Interoperability is centered on the Arbitrum ecosystem

Arbitrum's chain stack (the program formerly branded Orbit, now 'Arbitrum chains') lets you launch an L2 or L3 on Nitro, the most heavily used optimistic technology, now with BOLD permissionless fraud proofs. It is highly customizable: rollup mode or the cheaper AnyTrust (data-availability-committee) mode, a custom gas token, and custom DA. Note that popular Orbit chains like ApeChain and XAI run in AnyTrust mode, which trades some security for cost.

The trade-off unique to Arbitrum is commercial. Under the Arbitrum Expansion Program (AEP), a chain deployed outside Arbitrum One and Nova owes 10% of its protocol net revenue to the Arbitrum Foundation (split 8% to the DAO treasury and 2% to the developer guild), with a per-transaction alternative. OP Stack and the ZK stacks impose no such fee. If Nitro's maturity and the Arbitrum ecosystem fit, and the revenue terms work for your model, it is a strong optimistic choice.

Polygon CDK

Managed ZK chains on the AggLayer

Strengths

+ZK validity proofs give faster trustless withdrawals than optimistic stacks
+The AggLayer aims for cross-chain composability and shared liquidity
+Enterprise deployments at scale (OKX's X Layer, Immutable zkEVM)

Trade-offs

−Repositioned in 2026 to a managed chain-as-a-service; you work with Polygon Labs, not a self-serve kit
−AggLayer atomic composability is still maturing
−Do not confuse it with Polygon's public zkEVM chain, which is being sunset in 2026

Polygon CDK builds ZK-powered chains that connect through the AggLayer, Polygon's interoperability protocol, so assets and data can move between chains without routing back through Ethereum each time. The unified bridge is live; full atomic cross-chain composability is still maturing, so treat "shared liquidity" as a direction of travel rather than a finished feature.

The important 2026 change: Polygon repositioned CDK from a self-serve open-source toolkit into a managed chain-as-a-service, where you work directly with Polygon Labs to design and operate a bespoke chain. Real deployments include OKX's X Layer and Immutable zkEVM. Do not confuse CDK custom chains with Polygon's own public zkEVM chain, a separate product that is being sunset in 2026; that shutdown does not affect CDK chains.

ZK Stack (ZKsync)

Sovereign ZK chains on the Elastic Network

Strengths

+Sovereign chains: your own DA, sequencer, and governance
+ZK proofs, native account abstraction, and the Atlas upgrade's fast finality
+Chains join the Elastic Network, sharing liquidity and security via proofs, not bridges

Trade-offs

−EVM-compatible, not EVM-equivalent: a custom compiler introduces deploy-time gotchas
−Terminology has churned (Hyperchains to Elastic Network; the brand is now 'ZKsync')
−Smaller RaaS support than OP Stack

ZK Stack is Matter Labs' modular framework for sovereign, ZK-powered chains: a chain can bring its own data-availability layer, sequencer, and governance, and it has native, protocol-level account abstraction. The October 2025 Atlas upgrade added fast finality and higher throughput. One important caveat: ZK Stack is EVM-compatible, not EVM-equivalent, it compiles through a custom zkEVM compiler, so some contracts (different CREATE2 addresses, a few opcode and precompile differences) need attention at deploy time.

Connected ZK Stack chains form the Elastic Network (renamed from Hyperchains after a trademark collision), which shares liquidity and security through cryptographic proofs rather than bridges. For context, the ecosystem retired ZKsync Lite, one of Ethereum's first ZK rollups, in May 2026, consolidating on ZK Stack.

Verdict

Which should you use?

If your priority is

A general-purpose EVM L2 with the biggest ecosystem

you want the most-adopted, EVM-equivalent optimistic stack, Superchain interoperability, and the deepest tooling and talent pool.

→OP Stack

If your priority is

The most mature fraud-proof tech

you want Nitro and BOLD, the rollup-or-AnyTrust choice, and the 10% AEP revenue share fits your model.

→Arbitrum (Orbit)

If your priority is

A ZK chain you do not want to operate yourself

you want ZK withdrawals and AggLayer connectivity delivered as a managed service by Polygon Labs.

→Polygon CDK

If your priority is

A sovereign, deeply customized ZK chain

you want your own DA, sequencer, and governance, protocol-native account abstraction, and can handle EVM-compatibility (not equivalence).

→ZK Stack

Also consider

Other decisions that matter as much

Starknet's SN Stack (Cairo, STARK-based): the main non-EVM rollup stack, used to launch Cairo appchains and L3s (via Madara); it sits outside this EVM-focused comparison, but weigh it if you want STARK proofs and can build in Cairo instead of Solidity.
Data availability (Ethereum blobs, Celestia, EigenDA, Avail): since EIP-4844 blobs made Ethereum DA far cheaper, this is the single biggest lever on cost and security. Alt-DA layers cut cost further at a different security profile; choose deliberately.
Rollup-as-a-Service (Caldera, Conduit, AltLayer, Gelato, Ankr): these operate the base chain for you on these same frameworks (Caldera supports five stacks and ships a cross-chain Metalayer; AltLayer adds EigenLayer-based restaked security). You still need the explorer, oracles, indexing, and apps built on top, which is the layer we build.
Sequencer decentralization: all four default to a single sequencer. If shared or decentralized sequencing (Espresso, or AVS-based sequencing) matters to you, plan for it early rather than retrofitting it.

FAQ

What is the difference between an optimistic and a ZK rollup?

An optimistic rollup (OP Stack, Arbitrum) assumes transactions are valid and allows a challenge window, about seven days, during which a fraud proof can be submitted; native withdrawals wait out that window. Both leading optimistic stacks now run permissionless fault proofs and sit at Stage 1 on L2Beat. A ZK rollup (Polygon CDK, ZK Stack) posts a validity proof per batch, so it is proven correct and gives faster trustless withdrawals, at the cost of heavier proving infrastructure. For the largest EVM ecosystems the optimistic stacks lead; where fast withdrawals or the strongest proof story matter, the ZK stacks do.

Is Polygon CDK still an open-source, self-serve toolkit?

Not in the way it started. In 2026 Polygon repositioned CDK into a managed chain-as-a-service, where you work directly with Polygon Labs to build and operate a bespoke ZK chain on the AggLayer, rather than assembling it yourself. Real deployments include OKX's X Layer and Immutable zkEVM. Note two separate things share the Polygon name: CDK custom chains are active, while Polygon's own public zkEVM chain is being sunset in 2026.

What is the difference between ZK Stack and the Elastic Network?

ZK Stack is the framework you use to build a ZK chain; the Elastic Network is the interconnected system those chains form, sharing liquidity and security through proofs instead of bridges. It was previously called Hyperchains and renamed after a trademark collision, so older articles use the earlier name. Build with ZK Stack and opt in, and your chain becomes part of the Elastic Network.

Do I need a rollup, an appchain, or an L3?

A general-purpose rollup (L2) suits a chain meant to host many applications. An appchain is tuned to a single product, with a custom gas token, execution behavior, and only the infrastructure that product needs. An L3 settles to an L2 rather than to Ethereum directly, which can lower costs at the price of another dependency layer. The right answer depends on whether you need your own block space and fee model, and whether you are prepared to own a chain rather than deploy a contract. We help make that call before committing to a stack.

Do I have to run the chain myself, or can I use a Rollup-as-a-Service?

You can use a RaaS provider (Caldera, Conduit, AltLayer, Gelato, Ankr) to deploy and operate the base chain on any of these frameworks without running the node infrastructure yourself. That handles the chain layer, but not the surrounding layer a chain needs to be usable: the block explorer, oracle feeds, data indexing, Safe smart-account infrastructure, bridges, and the first applications. That surrounding layer is what we build and operate, whether you run the base yourself or through a RaaS.

Reviewed by Luis Medeiros, Field CTO at Protofire. Last updated: July 2026.

Build it

We build the layer that makes a new chain usable: the explorer, oracles, indexing, bridges, wallets, and the apps on top, on any of these stacks.

rollup and appchain development →

Book a call with Alejandro Losa

Launching a rollup or appchain? Talk to us about the stack choice and the infrastructure layer that makes it usable.