MEV on Layer 2: How Arbitrum, Base and Optimism Order Transactions

By Merunas Grincalaitis, founder of MEV DAO and a developer who builds MEV, sniping and trading bots on Ethereum. Last updated: Oct 31, 2026.

TL;DR: MEV still exists on Layer 2s, but it works differently than on Ethereum mainnet. Most rollups have a single sequencer and a private mempool, so classic sandwich attacks from the public mempool are much harder, while arbitrage, backrunning and liquidations move to latency races and fee auctions. Arbitrum sells a 60-second "express lane" through its Timeboost auction, Base and OP Mainnet order by priority fee (Base in 200 ms Flashblocks), and Flashbots measured spam bots using more than 50% of gas on several OP Stack rollups while paying under 10% of fees.

When people ask me to build "the same bot, but on Base" or "the same bot, but on Arbitrum", the honest answer is that it's rarely the same bot. The contracts and the ABI calls carry over. The part that decides who wins, transaction ordering, is completely different. This guide explains how MEV works on the big Ethereum L2s today and what that means if you trade, snipe or build bots there.

Is there MEV on Layer 2s?

Yes. Wherever there are DEX prices to rebalance, loans to liquidate and token launches to buy, there is MEV. What changes on an L2 is who decides the order of transactions and how you compete for a good position.

On Ethereum mainnet, MEV (maximal extractable value, the profit available from ordering, including or excluding transactions in a block) is mostly fought over through MEV-Boost: searchers send bundles to block builders, builders bid for the slot, and the validator takes the best block. On a rollup there is usually one operator, the sequencer, which receives transactions, orders them and produces L2 blocks before posting the data to Ethereum. Whatever ordering rule that sequencer follows is, in practice, the MEV market.

How is L2 MEV different from Ethereum mainnet MEV?

The main differences are a private mempool, very short block times and ordering rules set by one operator. Here is how the three largest Ethereum L2s compare, based on their own docs:

	Ethereum mainnet	Arbitrum One	Base	OP Mainnet
Who orders transactions	Builders via MEV-Boost, chosen by validators	Single sequencer	Single sequencer (base-builder)	Single sequencer
Public mempool	Yes	No	No public mempool to read	Private mempool
Block / ordering interval	12 s slots	250 ms blocks	2 s blocks, 200 ms Flashblocks	2 s blocks
Ordering rule	Highest-paying bundles and tips	First come, first served, plus Timeboost express lane	Priority fee within each 200 ms Flashblock	Priority fee
How bots compete	Bundles and builder tips	Latency and express lane bids	Priority fees and spam	Priority fees and spam

Sources: the Arbitrum Timeboost introduction, Base's transaction ordering docs and the OP Stack protocol overview and fee docs.

The practical result: on mainnet, a bot can see your swap in the public mempool and wrap a bundle around it. On these L2s, nobody except the sequencer sees your transaction before it's ordered, so the bot has to react to what already landed, not to what's pending.

Can you get sandwiched on Arbitrum, Base or Optimism?

Classic mempool sandwiching is much harder on these chains, but you can still lose value to bad slippage settings. A sandwich attack needs two things: seeing your trade before it executes, and getting one transaction in front of it and one behind it. Without a public mempool, the first part is gone for outside bots.

The OP Stack docs describe the sequencer's mempool as private, specifically to avoid opening up MEV opportunities. On Arbitrum, even the Timeboost express lane controller can only get its own transactions sequenced faster; it can't reorder anyone else's.

That doesn't mean you can set 20% slippage and relax. A loose slippage limit is still a gift to anyone who moves the price right before you, by accident or not, and the sequencer itself is a trusted party. I'd still keep slippage tight on L2s, and if you trade on mainnet too, keep using a private RPC there. Our guide to MEV protection covers that side.

How does Arbitrum Timeboost work?

Timeboost auctions a 60-second "express lane" whose transactions skip a 200 ms delay that everyone else gets. It went live on Arbitrum One and Nova on April 17, 2025, according to Arbitrum's Timeboost docs.

Before Timeboost, Arbitrum ordered transactions first come, first served. That turned MEV into a pure latency race: whoever had the fastest connection to the sequencer won the arbitrage, which pushed bots to spam and colocate. Timeboost keeps the 250 ms block time but adds a paid fast lane:

Every round lasts 60 seconds by default.
Before each round, bidders place sealed bids in a second-price auction run by an autonomous auctioneer. Bidding closes 15 seconds before the round starts.
The winner becomes the express lane controller and can send transactions to a special endpoint, timeboost_sendExpressLaneTransaction, which sequences them immediately.
Every other transaction gets its arrival timestamp pushed back by 200 ms.
The auction proceeds go to whoever the chain owner chooses, instead of all the MEV going to the fastest searcher.

The docs are clear about the limits. The controller can't reorder other people's transactions, isn't guaranteed the top of the block and isn't guaranteed a profit. The how-to page lists the Arbitrum One auction contract and auctioneer endpoint if you want to bid. Pricing a bid is the hard part: you're paying for a 200 ms edge over a whole minute, not for one known opportunity.

How does transaction ordering work on Base?

Base orders by priority fee inside 200 ms Flashblocks, so on Base you mostly compete with tips and timing. Base still produces a 2-second block, but its builder streams it in ten 200 ms pieces called Flashblocks, built with Flashbots.

Three details from Base's docs matter for bot builders:

Ordering locks every 200 ms. Once a Flashblock is broadcast, a later transaction with a higher priority fee can't jump into it.
Gas unlocks gradually. Flashblock 1 can use 1/10 of the block's gas, Flashblock 2 up to 2/10, and so on up to the full limit in Flashblock 10.
The mempool is dynamic. Transactions are sorted by fee when they're picked, not across the whole 200 ms window, so a high-fee transaction that arrives late can land after a lower-fee one. In other words, when your transaction arrives matters about as much as how much you tip.

For a sniper or arbitrage bot that means two levers: a competitive priority fee and getting your transaction to the builder early in the window. The Flashblocks WebSocket also lets you see preconfirmed transactions about every 200 ms, which is the closest thing Base has to a mempool feed.

What about OP Mainnet and other OP Stack chains?

OP Mainnet uses a private mempool and orders transactions by priority fee, highest first. The OP fee docs say the sequencer "will prioritize transactions with a higher priority fee and execute them before any transactions with a lower priority fee."

There was one experiment worth knowing about. In 2026 OP Mainnet ran a time-boxed test of stake-based priority ordering, where staking OP could improve your position, capped at a 3x advantage. The notice says it was limited to four weeks, after which ordering returned to standard priority-fee ordering. If you build on OP Stack chains, it's a reminder that the operator can change the ordering rules, and your bot's edge can disappear overnight.

Why do MEV bots spam L2s?

Because blockspace is cheap and there's no bundle market, so bots send many speculative transactions instead of one precise one. On mainnet, a searcher simulates offchain and only pays when a bundle lands. On a cheap L2 with priority-fee ordering, it can be more profitable to fire transactions that check prices onchain and do nothing if there's no opportunity.

Flashbots measured this in MEV and the Limits of Scaling. They found that:

Spam bots on several OP Stack rollups used more than 50% of gas while paying less than 10% of fees.
On OP Mainnet, spam used about 57% of gas but paid about 9% of fees.
Across a million Base blocks in February 2025, spam was about 56% of gas, 26% of L1 data usage and 14% of fees.
Two searchers were responsible for more than 80% of spam on Base.

They defined spam as transactions that make repeated DEX queries but transfer no tokens. A 2026 paper, To Wait or to Probe, studied Base arbitrage from June 2025 to February 2026 and found that this "probabilistic" style was 23% of arbitrage activity but produced 95% of spam, and that Flashblocks and higher fee floors pushed the market back toward targeted bots.

How should you build a bot for an L2?

Design around the chain's ordering rule first, then port your strategy. Here's the checklist I use when a client wants an L2 version of a mainnet bot:

Read the sequencer docs before writing code. First come, first served, express lane auctions and priority-fee Flashblocks each reward different things.
Measure latency to the sequencer endpoint. On Arbitrum without the express lane, and on Base within a 200 ms window, milliseconds decide the winner. Host near the sequencer's RPC and use WebSockets.
Drop mempool-based logic. Your triggers are new blocks, Flashblocks or event logs, not pending transactions. Backrunning landed state is the main play.
Budget fees per attempt, not per win. If you probe onchain, you pay for every miss. Track gas spent against profit captured, per strategy, every day.
Test on a fork. The same Anvil mainnet-fork workflow works against an L2 RPC, and it's the cheapest place to find bugs.
Plan for rule changes. Ordering policies on L2s are set by one operator and have changed several times in two years. Keep the competitive logic in a module you can swap.

Frequently asked questions

Is there MEV on Arbitrum?

Yes. Arbitrage, liquidations and backruns happen on Arbitrum. Since April 2025, part of that MEV is captured through Timeboost, a 60-second express lane auction, while other transactions face a 200 ms delay.

Does Base have a public mempool?

No. Base doesn't expose a public mempool to read. Its builder orders transactions by priority fee in 200 ms Flashblocks, and the closest thing to a live feed is the Flashblocks stream of preconfirmed transactions.

Can you get sandwiched on a Layer 2?

It's much harder than on Ethereum mainnet, because outside bots can't see your pending transaction in a private mempool. Keep slippage tight anyway, since the sequencer is a trusted party and loose limits still leak value.

What is Timeboost on Arbitrum?

Timeboost is Arbitrum's transaction ordering policy. A sealed-bid, second-price auction sells control of an express lane for each 60-second round, and the winner's transactions are sequenced immediately while everyone else's get a 200 ms delay.

Why are L2 MEV bots called spam bots?

Because many of them send transactions that check DEX prices onchain and usually do nothing. Flashbots found such bots used over 50% of gas on several OP Stack rollups while paying under 10% of fees.

Is MEV on L2s more profitable than on mainnet?

Not necessarily. Fees are lower, but competition is fierce, ordering rules change, and probing strategies pay for every miss. Profitability depends on your latency, your fee strategy and the specific chain.

Most of our work at MEV DAO is still on Ethereum mainnet, but more clients now ask for Base and Arbitrum versions of their strategies. If you're weighing that move, here's how our custom MEV bot development works.