Gas efficiency

Ekubo is designed to make trading cheap: one singleton contract holds every pool, and flash accounting settles only the net difference at the end of a transaction. This page measures what that means in practice — what a swap costs, how many swaps fit in a block, and what it would mean chain-wide — against Uniswap v4 and Uniswap v3. Every figure runs the deployed production contracts of all three protocols on a fork of the latest mainnet block, with fresh plain ERC20 tokens so token quirks do not blur the comparison. Harnesses, raw snapshots, exact calldata and the full method live in benchmarks/ in the docs repository.

Bottom line

One swap of 1 token through a single 0.3% pool of a fresh ERC20 pair, warmed, moving the price by at most one tick, at block 25999935 (September 17, 2026). Transaction gas is execution plus the 21,000 base plus the measured calldata:

Transaction, 1 pool	Ekubo	Uniswap v4	Uniswap v3
Production router (Yul router; Universal Router with Permit2; SwapRouter)	110,875	144,390	123,587
Minimal single-lock router on every side (favors Uniswap)	112,997	131,284	117,006
Native ETH in (Yul router; minimal locker; SwapRouter wrapping in flight)	96,496	115,892	126,792

Through the production routers, Ekubo’s transaction costs 23.2% less than v4’s and 10.3% less than v3’s. With every router stripped to the same minimal lock-swap-settle shape the gap is 13.9% to v4 and 3.4% to v3: at one pool v3’s pool is nearly as cheap, and the difference is in the router and beyond the first pool. With native ETH in, Ekubo is 16.7% below v4 and 23.9% below v3, which must wrap. All legs receive the same output within 0.005%, so the gap is overhead, not price. At the expensive end, 1000 USDT to USDC on the live mainnet pools (proxy tokens, blocklists) costs 133,750 on Ekubo against 155,960 on v4 and 151,515 on v3, 14.2% and 11.7% less: costly tokens compress the gap but do not close it.

Throughput per block

Mainnet’s block gas limit is 60 million. Real swap flow is routes, not single pools: every extra pool costs Ekubo about 24k of transaction gas against v4’s 36k and v3’s 59k, so the gap widens with route length. Dividing the limit by measured route costs gives each protocol’s ceiling on routes per block if a block held nothing else (minimal routers on every side, which favors Uniswap):

Routes per 60M block, if a block held nothing else	Ekubo	Uniswap v4	Uniswap v3
1 pool, production routers	110,875 → 541	144,390 → 415	123,587 → 485
1 pool, minimal routers	112,997 → 530	131,284 → 457	117,006 → 512
2 pools, minimal routers	137,091 → 437	167,027 → 359	176,438 → 340
3 pools, minimal routers	161,184 → 372	202,770 → 295	235,870 → 254

At three pools an Ekubo block holds 26% more routes than a v4 block and 46% more than a v3 block.

What this means for the chain

Pools per swap, measured over a month. Every 10th mainnet block from August 18, 2026, 18:36 UTC to September 17, 2026, 21:08 UTC (21,600 blocks, 25783810 to 25999800) was read through public JSON-RPC, and every successful transaction that emitted a pool-level swap event was counted by the number of such events. The event set covers Uniswap v2 and its forks, v3, v4, PancakeSwap v3, Curve (all generations), Balancer v2 and v3, Maverick v1 and v2, Fluid, DODO, Bancor v2.1, v3 and Carbon, Solidly-style pairs, and Ekubo:

Pools touched in the transaction	Share of swap transactions
1	69.1%
2	16.0%
3	7.4%
4 or more	7.5%

That is 664,726 swap transactions at an average of 1.75 pools per swap (daily averages 1.64 to 1.87; a 1,000-block window on September 17 gave 1.68). Those transactions used 32.7% of all gas in the sampled blocks (daily 20% to 41%). Transactions sent straight to the Uniswap and 1inch routers average 1.22 pools; the rest is aggregators, MEV and other contracts. Failed transactions emit no logs and are excluded by construction. Both figures are lower bounds: the count only sees swap events in the catalogued families, so any AMM outside the catalogue adds uncounted transactions and uncounted pools, and the swap-transaction gas share can only be higher. The chain-wide capacity gains below inherit that bias and are underestimates.

A swap’s transaction gas at p pools is the one-pool minimal-router transaction plus (p − 1) times the measured marginal per extra pool:

Average pools per swap	Ekubo	Uniswap v4	Uniswap v3	Ekubo saving vs v4 / v3
1.75 (measured)	131,019	158,020	161,461	17.1% / 18.9%
1.0	112,997	131,284	117,006	13.9% / 3.4%
1.5	125,044	149,156	146,722	16.2% / 14.8%
2.0	137,091	167,027	176,438	17.9% / 22.3%

Block-space shares: over the 30 days ending September 17, 2026, mainnet burned 1,255.7 ETH of base fees (ultrasound.money), and transactions to known DEX routers were about 4.2% of it, a lower bound that misses aggregators, MEV and direct pool calls; ultrasound’s coarser “defi” category was 19.4%. The 32.7% measured above counts every transaction that touched a pool, including MEV and aggregator flow whose gas is mostly not swapping. The 70% share was stipulated for this page as an “all AMM trading” upside case; it is an assumption, not a measurement. If a swap’s share of the burn equals its share of gas and the saving applies to all of it, moving that slice to Ekubo frees the share times the saving (each cell spans the saving against v3 and against v4):

Block-space share given to swaps	Measured 1.75 pools (v3 18.9%, v4 17.1%)	1 pool, production routers (v3 10.3%, v4 23.2%)	2.0 pools (v3 22.3%, v4 17.9%)
4.2%, measured DEX-router burn, lower bound	0.7–0.8%	0.4–1.0%	0.8–0.9%
19.4%, “defi” category bracket	3.3–3.7%	2.0–4.5%	3.5–4.3%
32.7%, swap-event transactions, measured	5.6–6.2%	3.4–7.6%	5.9–7.3%
70%, stipulated AMM-trading upside (not measured)	12.0–13.2%	7.2–16.2%	12.5–15.6%

Induced demand would fill freed space. Read these as an order of magnitude, not a forecast.

On Base and Arbitrum

L2 fees are cheap L2 execution plus a charge for the L1 data the sequencer posts. Both were measured on Base (block 51439900) and Arbitrum One (block 506178800) for 100 USDC to WETH through each protocol’s deployed contracts, warmed, on the deepest live Uniswap WETH/USDC pool of each chain; no live Ekubo pool there could absorb the swap, so the Ekubo leg runs on a pool created on the fork with the v3 pool’s price, fee and liquidity. The L1 data component is what each chain’s own fee oracle charges for the exact transaction, in gas at the block’s L2 base fee:

Chain	Leg (100 USDC to WETH)	Execution gas	Calldata bytes	L1 data (gas-equivalent)	Total, incl. 21,000 base and calldata gas
Base	Ekubo, Yul router	105,084	184	331	128,171
Base	Uniswap v3, SwapRouter02	121,803	228	319	144,646
Base	Uniswap v4, minimal locker	126,959	292	367	150,694
Arbitrum	Ekubo, Yul router	113,121	184	561	136,870
Arbitrum	Uniswap v3, SwapRouter	130,896	260	603	154,415
Arbitrum	Uniswap v4, minimal locker	135,006	292	644	159,234

The mainnet pattern carries over: the whole transaction is 14.9% and 11.4% cheaper than v4 and v3 on Base, 14.0% and 11.4% on Arbitrum, since both chains run the same EVM at the same opcode prices. The data component decides nothing at these blocks: 0.2% to 0.4% of every leg, and on Base not even smallest for Ekubo, whose dense route compresses worse than Uniswap’s zero-padded calldata. Compact calldata buys nothing measurable while L1 data is this cheap.

Why it is cheaper

• Cheaper pool core. One packed storage word and Q64 fixed-point math cost about 20k gas per swap, against 29k for v4’s multi-slot state and 37k for v3’s slot0, fee-growth and tick bookkeeping (lab flamegraphs in benchmarks/).
• Net settlement. Ekubo and v4 settle once per transaction (2 token transfers for a route of any length); v3 pays 2 per pool. On the production contracts each extra pool costs about 22k of execution on Ekubo, 34k on v4, 59k on v3.
• Compact calldata. The Yul router packs a single-pool route into 184 bytes; Uniswap’s Universal Router needs 1,092 for a single v4 swap with its Permit2 funding, v3’s SwapRouter 260.

Measured honestly

Providing liquidity is measured where providers actually touch it: production position managers on the live USDC/USDT pools. Minting a new ~10,000 USDC/USDT position costs 244,513 on Ekubo against 297,390 on v4 (17.8% less) and 381,653 on v3 (35.9% less). One-time costs (pool creation, tick initialization, first-swap accumulator writes) are excluded by design everywhere on this page, and the v4 Universal Router was measured only at one pool. Scope: EVM only, concentrated pools, no hooks or extensions on any side — any v4 hook or Ekubo extension changes every number here.

How it was measured

• Production Ekubo Core, Positions and Yul router; Uniswap v3 factory and SwapRouter; v4 PoolManager and Universal Router, on a fork of the latest mainnet block at run time, four fresh ERC20 tokens, 0.3% pools of matched liquidity, no tick crossings.
• Warmed, isolated steady state: one identical unmeasured call first, every touched slot already nonzero, each measured call its own transaction, so the snapshot is execution gas net of refunds; exact calldata recorded per call.
• Minimal single-lock routers of the same shape on all sides for route scaling, which favors Uniswap: its production routers cost more than the minimal figures, Ekubo’s costs less.
• L2s: Base and Arbitrum One forks pinned by block, deployed Uniswap routers and pools, the deployed Ekubo Yul router on a fresh pool mirroring the v3 pool; L1 data read from each chain’s fee oracle, not modeled.
• Pools per swap: a uniform 10% sample of a month of mainnet block receipts over public JSON-RPC, every event signature hashed from its protocol’s source. Not counted: order-book and RFQ fills unless they land in a pool, the legacy Ekubo v2 core, long-tail AMMs. Sources, snapshots, calldata, pool identifiers and reproduction commands: benchmarks/ in the docs repository.