Rise of the arb bots! How arbitrage is driving Hedera's transaction volume

On 10 August 2026 a Hedera account that had sat idle for over a year received 2,200 HBAR. Two hours later it made its first trade through SaucerSwap, Hedera's biggest decentralised exchange. Three months on, that account and the contracts it has deployed, which name themselves “cyclebot” on-chain, make almost a third of all transactions on the network.

Cyclebot is an arbitrage bot and it isn't alone on the Hedera network. A second bot operated by OnlyFlies started eleven days later. By November 7 the two of them accounted for 58% of transactions on Hedera that day. If you've noticed Hedera's TPS climbing since August, these bots (along with a handful of smaller arb bots) explain most of that growth.

This report looks at what the bots do, how much of the network they account for, and why so few of their transactions are trades. That last part says a lot about how Hedera differs from other crypto networks like Ethereum or Solana.

01What an arbitrage bot does

A decentralised exchange like SaucerSwap prices each pair of tokens in their own pool. When someone makes a big trade in one pool, the prices of its tokens change while prices in other pools remain unchanged. For a brief moment, an arbitrage opportunity may arise in which an indirect trade looping through several liquidity pools may return more value than the bot initially started with, allowing the bot to complete the trade and make a small profit.

This particular form of arbitrage is known as cyclic arbitrage, and it's what the largest arbitrage bots on Hedera are doing. These arbitrage bots watch for such opportunities and attempt to trade through the loop before the price discrepancy closes or anyone else beats them to the punch. Each trade nudges the pools back into line, which is why exchanges generally consider arbitrage healthy.

Since early October the largest arbitrage bot on Hedera, Cyclebot, has actually taken this arbitrage strategy one step further and its main contract is no longer even using its own money to trade. Instead, Cyclebot now borrows from the first pool, trades through the loop and repays the loan in the same transaction. This is called a flash swap, it requires no collateral for the loan. If the numbers no longer work by the time the transaction runs, the whole thing simply reverts and the bot just pays a small fee. Figure 1 shows one real trade from 9 November.

That trade made 2.49 HBAR, about 17 cents, for a 0.37 HBAR fee. That seems like an inconsequential profit, but Cyclebot made similar trades about 360 times a day over the last month, and occasionally it fared much better: for example, one trade through an HBAR/PACK pool made 111.7 HBAR in profit, minus a fee of just 0.33 HBAR (about 2 cents).

02Getting to know Cyclebot

Cyclebot has no website and no public owner that we know of. Its contracts name themselves in their memos (“cyclebot executor v2 proxy”, “cyclebot mixed V1-V2 v1.4.0”), and every one of them was deployed by one of its two operator accounts. All of Cyclebot's trades include a memo naming the route it's attempting and a tag – until late August that tag was simply “Cyclebot V2”; since then it's been “GRELF MAFIA”, perhaps providing a clue about ownership. “GRELF MAFIA” is what the community around the GRELF meme token calls itself (grelf.me).

Since becoming active on the network in early August, Cyclebot has changed fast:

• 10 August: trading through SaucerSwap's router. The operator account, 0.0.8049585, was created in January 2025 and then barely used. It received 2,200 HBAR from 0.0.108869 and made its first swap through SaucerSwap's public V1 router the same afternoon, using its own HBAR.
• 28–29 August: its own contracts. v1.4.0 and then v1.5.0, both trading across SaucerSwap's V1 and V2 pools. Activity grew through September and jumped on 28 September, its first day above 250,000 records.
• 3 October: flash swaps. A new executor behind an upgradeable proxy, 0.0.10903897, which borrows from the first pool instead of using Cyclebot's money. Four days later a batch executor followed. Six versions of the executor's code have been deployed since.
• 13 October: a second operator. 0.0.509985, an account from 2021, got 800 HBAR from the first operator and nine minutes later deployed a second copy of the contracts. Since 27 October it has done one job: pinging nodes, which we explain in section 04.

• Account Main operator 0.0.8049585

  Signs and pays for every trade. Deployed the original contracts. Active since 10 August 2026.

• Account Second operator 0.0.509985

  Funded by the main operator. Sends the node pings, around 200,000 a day.

• Contract Executor v2 (proxy) 0.0.10903897

  Does the flash-swap trades and keeps the profit as WHBAR. Most of Cyclebot's trades go through it.

• Contract Batch executor 0.0.10909136

  Tries several routes in one transaction. Deployed 7 October 2026.

03What's the network impact?

Hedera's transaction count, the figure behind its published TPS, counts every record the network writes, including failed transactions and records generated inside a transaction. By that measure, the measure Hedera itself uses, the two bots went from almost nothing in August to 34% of all records in October and 46% so far in November.

Over the last 30 days Hedera averaged 8.3 TPS. Cyclebot accounted for 2.5 of those and OnlyFlies 0.9. In August the network averaged 3.7 TPS, so the rise since then is about 4.6 TPS, and the two bots account for about three-quarters of it.

In terms of network revenue the number is smaller but still meaningful. In the last 30 days the network took in about 200,000 HBAR (roughly US$15,500) in fees, and these two bots paid 19,200 HBAR of it (roughly US$1,500), or 9.6%. Cyclebot alone paid 7.2%.

04Why so many transactions?

If Cyclebot makes only about 360 successful trades a day, why does it make more than 200,000 transactions a day? Figure 3 breaks its activity down. Trades, successful or not, are only about 1% of them. The vast majority are node pings and duplicate transactions.

These transactions, which account for almost all of Cyclebot's activity, need a little more explanation, as they're critical to understanding how arbitrage works on Hedera and what bots need to do to gain an advantage.

05On Hedera arbitrage is all about speed

On most blockchains arbitrage is an auction, on Hedera it's a race. That difference explains everything in Figure 3, all those extra transactions happen to maximise speed when an arbitrage opportunity comes along.

So Hedera's bots spend their effort on speed. They send each trade through two dozen nodes to shave off the slowest path, ping nodes constantly to measure their speed, and fire at every possible opportunity because a miss is cheap. Arbitrage value still exists on Hedera, but you can't buy the opportunity to claim that value, there's no MEV auction like you see on Ethereum. Instead on Hedera if you want arbitrage value you have to race to claim the prize and that race results in a lot of transactions.

06Does it pay?

For Cyclebot, yes. We added up every HBAR and WHBAR that moved in or out of its 16 accounts and contracts from 1 August to 9 November (Figure 4). Its owner put in 3,200 HBAR and has since withdrawn 37,175 HBAR; on top of that it has paid 23,181 HBAR in fees and still holds 28,363 HBAR. Its trades brought in about 85,500 HBAR before fees and about 62,300 HBAR after, roughly US$5,200 at the HBAR prices of the day, over about three months of trading.

That's about 3.15 HBAR per successful trade on average, across 27,133 successful trades. The withdrawals went to six outside accounts, most of it to 0.0.1127968 (23,147 HBAR in four transfers) and 0.0.9214873 (11,405 HBAR). We haven't identified who's behind them.

07OnlyFlies, and the rest

The second bot belongs to OnlyFlies. Its main account, 0.0.10176974, started trading on 21 August and runs a family of contracts it has replaced every few weeks: an early version that held its WHBAR in the contract, then “skip-or-fill” and “sized-run” bots such as 0.0.10888630 and 0.0.10921760. Unlike Cyclebot, it trades with its own WHBAR rather than borrowing. A second account, 0.0.10937292, has sent its pings since 29 October.

In the last 30 days OnlyFlies made 2.2 million records, 10.4% of the network, and paid 4,756 HBAR in fees. It duplicates more and pings less than Cyclebot, and it hits less often: 2,397 successful trades out of about 36,000 tries.

They aren't the only ones. We've seen smaller accounts with the same signature, mostly reverts and the odd loop that pays, such as 0.0.8004788. None of them is large enough to change the picture, so we've left them out of the figures.

08What it means for Hedera

There's good news here, with a big caveat.

• Real, unsubsidised revenue. Nearly 10% of network fees now come from bots that pay their own way because the trade is worth it. Nobody granted this activity into being.
• Better prices on SaucerSwap. Every successful loop pulls the pools back into line, so the next person to swap gets a fairer price.
• Fair ordering, working as designed. The bots can't buy position, so they can't front-run or sandwich anyone. What they compete on is speed.
• The caveat: read TPS carefully. About 40% of Hedera's recent transaction records come from two bots, and most of those are pings and duplicate copies. The rise in TPS since August is mostly them, not new users. Fees tell a more measured story: the bots now pay about a tenth of Hedera's network fees, a significant share, but far less than their 40% of transactions, because pings cost very little and duplicates cost nothing. In US-dollar terms the network's daily fee revenue rose about 20% between August and October, and the bots supplied roughly 40% of that rise.

Arbitrage bots usually show up where there's trading worth arbitraging, so in that sense their recent growth on Hedera is a vote of confidence in the network's burgeoning DeFi ecosystem. When you see Hedera's headline TPS number, though, it can now be a little misleading because so many of those transactions are either node pings or duplicates which provide little real value – it's always worth knowing what the network activity actually is and who's behind it rather than just a single TPS number.

All in all this recent surge in arb bot activity has been good for the network, boosting both its TPS and revenue considerably. It'll be interesting to see if it continues to grow or if the space is already saturated given the difficulty of competing purely on speed.

09Verify our research yourself

Cyclebot on Hedera

• Operators: 0.0.8049585, 0.0.509985
• Contracts: v1.4.0, v1.5.0, executor v2 proxy, batch executor, WHBAR unwrapper, second proxy (13 Oct)
• Sample transactions: a successful trade (Figure 1), a reverted trade, a ping, the first funding, the largest withdrawal

OnlyFlies on Hedera

• Accounts: 0.0.10176974, 0.0.10937292
• Contracts: skip-or-fill bot 0.0.10888630, skip-or-fill bot 0.0.10904282, sized-run bot 0.0.10921760

The pools in Figure 1

• SaucerSwap V2 HBAR/USDC, V1 HBAR/SAUCE, V2 SAUCE/USDC

Network figures

• HashScan dashboard (network TPS)
• hgraph-io/hedera-stats, the open-source code behind Hedera's published transaction and fee figures