Invest1 publisher3 min readPublished
Jupiter Ultra scores 0.7 on sandwich exposure against Axiom's 18.9 in a six-chain study
Researchers counted 28 million Solana sandwich attacks in three years and found Jupiter Ultra users hit far less often than users of rival trading terminals. Because that spread sits between apps on the same chain, it is evidence about the choice of terminal more than about the choice of chain.
The Investor · Invest desk
What happened
- The paper, 'No Place to Hide', was posted to arXiv on September 23, 2026 by researchers from Category Labs, ETH Zurich, Flashbots and the University of Lisbon.
- Photon scored 11.1 on the same measure, and BullX and GMGN also came in above 10.
- Jupiter claims 34 times better sandwich protection than rivals, average positive slippage of 0.6 basis points and execution fees 8 to 10 times lower.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- cost On Solana, the worse execution a sandwich imposes lands mostly on users of terminals scoring above 10, so the terminal a trader picks decides how much of that cost they carry.
- decision A Solana trader trying to cut sandwich losses gets better support from this study for switching terminal than for leaving the chain.
- contradiction Jupiter's 34-fold protection claim is wider than any gap the study's ratios produce, so it cannot be cited as the researchers' finding.
- constraint With validators still an attack vector, app-level routing can cut sandwich exposure on Solana but cannot end it, whichever terminal a trader uses.
Divide Axiom's 18.9 by Jupiter Ultra's 0.7, the excess ratios Crypto Briefing reports from the paper [4] [6], and the gap between two apps on the same chain is 27 times [1]. Photon's 11.1 puts it near 16 [2]. If Axiom's number is a single-victim ratio, the matching comparison is Jupiter's 2.0 [5]. That gap is about 9.5 times [3], and it leaves Jupiter's own single-victim users sandwiched at twice the rate the study treats as expected [7].
Twenty-eight million attacks over three years [3] comes to roughly 25,600 a day [4]. Each works the same way: a bot sees a pending swap, buys ahead of it to push the price up, then sells after the swap fills at the inflated price, and the victim gets worse execution [11]. Crypto Briefing attributes Solana's volume to low fees that let bots try cheaply at scale, and to order visibility that shows them what to front-run [12]. The slippage tolerance a user sets caps how far a bot can move the price against them. Jupiter says Ultra V3 addresses both sides, with private routing that hides the transaction until execution and slippage estimates that adjust in real time in place of a static user-set figure [7].
Jupiter's attribution runs into timing. V3 went live on October 17, 2025 [7] and the paper reached arXiv on September 23, 2026 [1], so no more than about 11 months of a 36-month window, roughly a third, can reflect it [5]. The company's claim of 34 times better protection [8] is also wider than the largest gap the study's ratios produce, 27 times against Axiom [6].
The researchers' broader conclusion, as Crypto Briefing reports it, is that the chain is not the only variable and that app choices about routing, public exposure and slippage play a decisive role [9]. Crypto Briefing's account does not include excess ratios for apps on Ethereum, Tron, Base, Arbitrum or Monad [2], or a dollar figure for what the attacks took from traders. The spread between apps on Solana is the part a reader can check. Ranking app design above the chain rests on the paper's own reading.
Customer mix is one alternative explanation for the spread, and the V3 timing is another. If the terminals scoring above 10 draw traders who set wide slippage, the ratios measure the users as much as the routing. Validators limit any app's result: the researchers say they remain a vector, and that Jupiter Ultra reduces sandwiching on Solana without solving it [10]. Private routing hides an order until execution [7], and execution still passes through a validator.
I'd expect the direction to hold. A 0.7 against four rivals above 10 on one chain [6] is a wide gap for customer mix alone to produce. The view would be wrong if Jupiter Ultra's ratio after October 2025 proves no better than before it, or if Solana's best app turns out worse than a typical app on another chain.
What to watch
- Whether Axiom, Photon, BullX or GMGN add private routing or real-time slippage limits, and whether their excess ratios then fall toward Jupiter Ultra's.
- Whether sandwiching attributed to validators rises on Solana as private routing spreads, moving the extraction from bots to block producers.