The State of MEV on Solana

In plain terms#

What it is. Every few hundred milliseconds one Solana computer, the leader, decides which transactions go into the next block and in what order. That choice is worth money: being placed right before or right after a trade that moves a price lets you profit from the move. The value captured this way is called MEV, short for maximal extractable value.

The main kinds on Solana.

• Arbitrage: a big swap moves a token's price on one exchange, so a bot buys it there and sells it on another exchange that has not caught up. Solana saw about 90 million of these in 2024, averaging around $1.58 each.
• Liquidations: a borrower falls below their collateral limit and a bot closes the loan for a fee.
• Sandwiching: a bot sees a user's swap before it lands, buys ahead of it to push the price up, lets the user fill at the worse price, then sells into it. Sandwiching took between $370 million and $500 million from Solana users in the sixteen months to May 2025, and live trackers still counted thousands of SOL a day in September 2026.

A tiny example. A user buys $100 of a small token. A sandwich bot spots the order early, buys first so the price ticks up, lets the user buy at that higher price, then sells its own tokens into the user's purchase. The user gets a few cents to a few dollars less of the token than they should have; the bot pockets that gap minus what it paid to get placed. Tens of thousands of victims a day add up.

Who actually takes it. To be placed well you pay either a priority fee (a fee inside the transaction that goes to the leader) or a tip attached to a bundle, a small package of transactions that lands together or not at all. Arbitrage bots hand over half or more of their profit in tips because so many of them compete, so most of the money ends up with validators, the computers that produce blocks, and the people who stake SOL behind them. Even for them it is a small slice: about nine tenths of a staker's income is the network's own inflation, and tips have shrunk from their early-2025 peak to under $10 million a quarter. Among bots, a few well-funded teams win most races; one program did about half of all sandwiches in late 2024.

Why sandwiching is the contested one. Arbitrage and liquidations react to something that already happened and leave prices fairer than they found them. Sandwiching needs to see a user's trade before it lands and takes money straight from that user. Solana has no public waiting room for pending transactions (no mempool), so seeing a trade early means running the leader yourself, running altered software, or buying a private feed. Jito, which runs the biggest tip system, shut its preview feed in March 2024 over sandwiching, and the Solana Foundation removed more than thirty validators from its stake-support program for it. The newer block builders, BAM from Jito and Harmonic, refuse to build sandwiches on the roughly half of all blocks they now handle. It has not disappeared, but it is smaller and hits small trades.

What can go wrong. For users, a bad fill from a sandwich, though wallets now route most trades through protected lanes. For bots, most attempts fail and still cost a fee, the fast data feeds are now paid, and margins have fallen for five straight quarters. For the network, bots firing many attempts at once clog blocks, because failing is nearly free.

Terms you will meet on this page. A DEX is an exchange run by a program instead of a company. An LP (liquidity provider) deposits tokens into a DEX pool and earns fees; LPs are who arbitrage bots trade against, and their quiet loss to stale prices is called LVR. A slot is one block interval, about 250 milliseconds. Jito is the company whose software most validators run to sell block placement; BAM and Harmonic are competing ways of building blocks. Stake is the SOL locked behind a validator. A prop AMM is a private, oracle-priced pool run by one market maker; these now handle most trading in the big tokens.

Bottom line. MEV on Solana is a race to see the next block first and pay the right builder the right price. The benign kinds are a low-margin infrastructure business run by a few teams; sandwiching survives at the fringe. The money mostly flows to validators and stakers, and it shrinks every quarter.

TL;DR#

• Solana MEV is a latency and state-visibility game, not a mempool game. There is no public mempool, blocks stream continuously, and the leader is known in advance.
• Value flows to validators and stakers through two channels: Jito tips (an out-of-protocol bundle auction) and priority fees (in-protocol, 100% to the leader since SIMD-0096).
• The tip market has collapsed from its Q1 2025 peak: Jito tips were $9.9M in Q2 2026 vs $19.85M in Q1 2026 and roughly $767M for all of 2024. Priority fees are now 60% of Solana's real economic value; tips are under 20%.
• Sandwiching was never eliminated. It moved from Jito's mempool (shut March 2024) to private mempools and modified validators, was pushed to the fringe by delegation-program removals and Jito's blacklist, and persists at a lower level today. Live trackers still show thousands of SOL a day extracted in September 2026.
• Block building consolidated into three camps: Jito BAM (~33–34% stake), Harmonic (~21%), and Rakurai (~9%), on top of classic Jito-Agave. All three sell ordering to searchers under "no sandwiching" rules.
• The frontier moved upstream: who sees state first (DoubleZero shreds, BAM preconfirmations) and who pays the right builder the right price.

How value reaches validators: a priority fee is paid to the leader who includes the transaction (SIMD-0096, live February 2025, ended the 50% burn). A Jito tip is a SOL transfer to one of eight tip accounts inside a bundle; the leader sweeps it to a tip-distribution account and stakers claim their share at epoch end via a merkle root, net of validator commission and a Jito protocol cut. Harmonic has no tip account; its "tip" is the priority fee and goes 100% to the validator. Details of each path are in solana-block-building.md.

2. The extraction landscape in numbers#

Aggregate flows#

Jito's tip volume is the best public proxy for the paid-ordering market, and it has fallen for five consecutive quarters.

The transaction count through Jito did not fall; the price paid per transaction did. That is a margin story, not a volume story.

Solana's total REV (real economic value: priority fees + tips + vote and base fees) was $51.0M in Q2 2026, down 43% from $89.8M in Q1. Priority fees were $30.8M (60%), Jito tips $9.9M (19%), vote and base fees $10.3M. Galaxy Research puts total network fee revenue at ~$155M for the quarter (−44%) and notes that inflation still provides ~90% of staking yield, so MEV is a rounding error for the average staker today.

Share by strategy#

No single source measures every strategy on a consistent basis, so the table mixes windows. Treat it as order-of-magnitude.

Two figures circulate widely without a traceable primary source and should be treated as (unverified): "$720M of MEV revenue in the last year" and "$480M cumulative searcher profits through Jito bundles". Both appear on infrastructure vendors' blogs, not in Blockworks, Helius, or Ghost data.

Timeline

Did BAM and Harmonic end it? No, but they changed its shape. Both stacks refuse to build sandwiches on the ~54% of stake they cover, and TEE-based ordering in BAM removes the leader's own ability to peek. What remains is concentrated on the ~45% of slots led by plain Agave or unaligned Jito-Agave validators, and in the "blind" or "wide" variants that do not need adjacency: a bot statistically guesses that a Pump.fun-style buy is coming and brackets it across two transactions 50–300ms apart. That is lower-margin and hits small trades, which matches the live tracker's profile of many victims and low per-victim loss. There is no measurement yet of sandwich rate by client (Ghost has a "clients" view but no 2026 report), so the "BAM/Harmonic ended it" claim is plausible for their own slots and unproven for the network.

Protection products for users (what each actually does):

The common thread: every product works by keeping flow away from the open ordering market. That protects users and shrinks the pie for independent searchers at the same time.

4. Infrastructure and market structure#

Block builders#

Numbers come from Blockworks (Q2 2026) and SolanaFloor/Syndica (Mar 2026); categories overlap because Frankendancer can run BAM or Harmonic overlays. The "95% of stake runs jito-solana" figure often quoted counts code lineage, not who sells ordering: Rakurai and several forks descend from jito-solana.

Harmonic launched 5 Nov 2025 with a $6M Paradigm-led seed; co-founder Jakob Povsic also co-founded Temporal (Nozomi). Harmonic's pitch is that the validator, not the builder, chooses the ordering rule; its critique is that Jito is conflicted by owning infrastructure, BAM and stake-pool scoring. The same critique applies to Temporal, which runs a transaction-landing service and is widely believed to operate HumidiFi (unverified). Syndica's April 2026 data showed Harmonic's Firedancer "PRF" validators earning the most per block (0.0555 SOL, +79% vs mean), almost all of it priority fees rather than tips.

Data and preconfirmations#

• Jito ShredStream shut down on 5 Sep 2026 after a 60-day notice in early July. Jito pointed users to DoubleZero Edge, a paid private-fiber shred feed covering 62.6% of stake. On 9 Sep 2026 DoubleZero cut off RPCs that had been retransmitting its feed without a licence and claimed a 70%+ lead on leader shreds. A competing vendor's benchmark (OrbitFlare, Frankfurt) had already put Jito's feed 6.5ms behind its own at the median. Early state is now a licensed product.
• BAM preconfirmations went live 9 Sep 2026 via Helius and Triton: BAM-sequenced transactions are streamed to subscribers before the block, a claimed 5–10ms p50 edge over shreds. Revenue splits 35% to BAM validators (stake-weighted, paid as priority fees), 35% to the Jito DAO, 30% to distribution partners; payouts start October 2026. A preconfirmation is a schedule, not a guarantee; the BAM Verifier disconnects validators that deviate.

Together these two events mark the point where information about the next block became a priced good sold by the block builder and the network operator, rather than something a searcher could infer for free by running nodes.

Where the order flow went#

Two structural shifts matter more to a searcher than any block-builder change.

Proprietary AMMs. SolFi (Ellipsis), Tessera V (Wintermute), HumidiFi, ZeroFi, GoonFi and Obric quote from oracles, refresh quotes in as little as 143 CUs, buy cancel priority with small Jito tips, and reject takers below 100k CUs as likely bots. They took over 60% of SOL/USDC volume by July 2025 (peak 86%), roughly 65% of all on-chain volume by December 2025 (Chorus One), and quote 0.4–1.6bp on SOL/USDC. Nearly all their volume (88–99%) arrives through aggregators. The effect is that the stale-quote arbitrage that made up most of 2024's $143M of atomic-arb profit is now internalised by market makers who are never stale. Passive-pool LVR is still there on long-tail tokens; on majors it has largely been competed away.

Jupiter and order-flow internalisation. Jupiter routes over 80% of aggregator volume and over half of all DEX volume. Ultra v3 sends most of that through private lanes and an RFQ, so the backrun a searcher used to pick up after a large Jupiter swap increasingly does not exist on the public path. DFlow, Titan and OKX compete on the same model (DFlow reportedly topped daily aggregator share at one point in 2026; unverified). On the auction side, Pyth Express Relay runs protocol-controlled auctions for liquidations and swaps with a named searcher set (Wintermute, Flow Traders, Auros and others), and BAM plugins are the Jito-native path for an application to define its own ordering or auction and take a fee. Solana has no Flashbots-style MEV-Share; the equivalent role is being filled by aggregators and by builder plugins, each keeping the rebate inside its own product.

6. Trends and open questions, next 12 months#

Does BAM plus Harmonic remove the searcher, or relocate it? Relocate. Both stacks still need someone to find the backrun and pay for it; what they remove is the bundle-versus-banking-stage lock race and the leader's ability to self-deal. The searcher role migrates toward three places: plugin authors who sit inside the application's ordering rule, market makers who quote on prop AMMs and RFQs, and preconfirmation subscribers who act 5–10ms before the block. The independent bundle-submitting searcher of 2024 is the party being squeezed.

The preconfirmation information race. Preconfs turn the sequence into a paid feed. Expect: subscriber concentration among a handful of latency shops; a tension between BAM's TEE privacy story and selling the schedule early; Harmonic answering with its own stream or arguing that FIFO makes preconfs worthless. Watch the October 2026 payout numbers to see how much validators earn from selling early sight versus from tips.

Encrypted and TEE ordering. BAM's TEE is the first production example; Harmonic's argument is that ordering-rule choice, not hardware trust, is the guarantee. Open questions are attestation transparency, what plugins can see, and whether TEE side channels become a searcher edge.

MEV under Alpenglow and shorter slots. 250ms slots and 150ms finality shrink every batch auction and leader window. Helius's own analysis expects leaders with in-house building to gain and independent latency arbitrageurs to lose. Vote transactions leave blocks, freeing compute. Multiple concurrent leaders, if they arrive, would split the write set and create cross-lane MEV that nobody has designed for.

Regulatory attention. The Peraire-Bueno Ethereum case ended in a mistrial in November 2025 with prosecutors seeking a retrial; it turned on exploiting MEV bots, not on sandwiching itself, and left the legality of sandwiching unresolved. ESMA published a risk analysis of MEV in July 2025. The SEC/CFTC March 2026 interpretive release classified SOL as a digital commodity and said nothing specific about MEV. Nothing today restricts benign extraction; sandwiching remains a reputational and delegation problem rather than a legal one, and that could change with one enforcement action.

Searcher margin compression. Tips per Jito transaction halved in one quarter while transaction count held flat. Prop AMMs removed most stale-quote arbitrage on majors. Aggregators internalise backruns. Shred and preconf data are now paid. Compute-priced fees (SIMD-0553) would have made probing dearer, but validators rejected it in August 2026. Every one of the others is a permanent, not cyclical, reduction in the share of DEX volume that reaches an open ordering market.

7. What this means for a searcher operation#

Where the remaining edge is

• Early state. DoubleZero Edge for shreds and BAM preconfirmations via Helius or Triton are now the only fast feeds; the free tier is gone. Budget for both and measure your own p50 against them.
• Per-builder tip calibration. Each leader is now a BAM, Harmonic (FBA/FIFO/MREV), Rakurai, or plain Agave slot with different price sensitivity. Bidding a Jito tip into a FIFO slot is wasted money; bidding a priority fee into a BAM slot without a bundle loses the lock race. See the sender playbook in solana-block-building.md.
• Benign strategies only. Backruns, cyclic arb on long-tail and launch tokens, liquidations, and JIT liquidity all run inside every builder's rules. Liquidations in particular are increasingly auctioned (Express Relay) or app-sequenced (BAM plugins); getting whitelisted where that is the model is part of the job.
• Long-tail pools. Passive LVR still exists where prop AMMs do not quote. That is where 2024-style stale-quote arb survives.

What has closed

• Mempool sandwiching (March 2024) and its private-mempool successors on any BAM or Harmonic slot; what remains is small, blind, and a delegation-program liability for any validator that hosts it.
• Spam as a substitute for latency: 250ms slots and tighter schedulers make blind probing lose money (compute-priced fees would have added to this, but SIMD-0553 was rejected).
• Free early sight (ShredStream) and majors arbitrage against passive pools.

What to watch

• Blockworks Q3 2026 Jito report: whether tips stabilise below $10M/quarter, and first BAM preconfirmation revenue.
• Ghost per-client sandwich rates for 2026; whether the residual is on plain Agave slots.
• Alpenglow mainnet activation (no date set as of 25 Sep 2026; mainnet feature activation resumes 28 Sep) and the 200ms slot gate.
• Whether a revised SIMD-0553 returns after its August 2026 rejection; SIMD-0550 and SIMD-0123 feature-gate activation.
• Harmonic's response to preconfirmations, and whether its stake share moves from 21%.
• Any Jupiter, DFlow or Titan move to a formal order-flow auction with searcher rebates, which would reopen internalised flow to outside bidders.