Many Layer-1 (L1) networks promote theoretical or peak TPS (transactions per second) figures that differ substantially from typical mainnet throughput. If you're picking a chain to build on, the gap between marketed capacity and live performance directly impacts user experience under load.
Note: Benchmark metrics were captured on July 16, 2026, at 12:00 UTC. Mainnet TPS, validator counts, and network metrics change continuously.
This article builds a comparison table using observed mainnet performance data, not lab benchmarks, and walks through why throughput, finality (the moment a transaction becomes irreversible), and decentralization pull against each other. Note that TPS definitions vary by network regarding the inclusion of consensus votes, failed transactions, and batched operations.
The Benchmark For Real World Performance And Decentralization
The table below combines mid-2026 chain-explorer data with the Nakamoto coefficient, a metric academics use to express the minimum number of entities that would need to collude to compromise a network. A higher number means an attacker needs to control more independent parties, so it's generally read as a decentralization signal, though it has real limits we cover below.
Blockchain | Actual TPS (observed) | Finality time | Validator count | Nakamoto coefficient |
Solana | 1,480 – 4,000 | ~12.8 sec | ~800 | ~20 |
Ethereum | 12 – 35 | ~12.8 min | ~10,000 nodes (1M+ validator keys) | High (stake-weighted) |
Sui | 47 – 500 | 480ms – 2 sec | 126 | 19 |
Aptos | 155 – 800 | 900ms | 97 | 14 |
Avalanche | 43 – 200 | < 1 sec | ~1,200 | Variable |
Cardano | 5 – 10 | ~5 min | ~3,000+ | 116 |
Bitcoin | 3 – 12 | ~60 min | N/A (miners) | 4 – 5 |
A few notes on where these numbers come from and where they're soft:
- Solana and Aptos figures are cross-checked against Chainspect's live comparison dashboard, which pulls directly from each chain's real-time block data rather than a static report. At the time of writing, Chainspect showed Sui at roughly 47 tx/s and Aptos at roughly 155 tx/s over a one-hour window, with Sui's max observed TPS (100-block sample) at 927 and Aptos at nearly 13,000. These numbers move constantly with network load, so treat the ranges above as a guide, not a live reading, and check the dashboard directly if you need a current number.
- Bitcoin's Nakamoto coefficient is commonly cited as 2-4 in older marketing material, but the peer-reviewed measurement study that actually computed it from a full year of block data found it fluctuating mostly between 4 and 5 across daily, weekly, and monthly windows in 2019. We're using that figure here because it's traceable to a real measurement methodology rather than a repeated but unsourced number. Mining pool concentration shifts over time, so this should be treated as a historical range rather than today's exact reading.
- Cardano's 116 reflects the chain's current "K" parameter setting of 500 stake pools. There's a live governance proposal to raise K to 1,000, which modeling from an Edinburgh University blockchain research group projects would nearly double the coefficient to around 226. That 226 figure is a projection tied to a proposal that has not passed as of this writing, not a current fact, so don't cite it as Cardano's present decentralization score.
Throughput Compared To Actual TPS
Every chain publishes a theoretical maximum measured under lab conditions: simple transfers, no contention, perfect parallelization. Actual mainnet TPS is almost always a small fraction of that number, for three reasons.
- Transaction complexity. A DeFi swap or NFT mint does far more computational work than a plain token transfer, so a network's real throughput depends heavily on what kind of transactions people are actually sending.
- State contention. Parallel execution engines like Solana's Sealevel or Sui's Move-based runtime speed things up by processing non-overlapping transactions simultaneously. But when many transactions try to touch the same piece of state (say, the same liquidity pool), they get serialized anyway, and the parallelism advantage shrinks.
- Network overhead. Propagating blocks and coordinating validator votes takes real time and bandwidth, and that overhead scales with validator count, not down.
This is why Aptos advertises a theoretical maximum of 160,000 TPS, but Chainspect's live data shows it processing closer to 150-800 transactions per second on mainnet. That gap is just what happens when lab conditions meet real usage.
The Finality Fallacy
For developers building anything that touches money, for financial applications, finality can be as important as, or more important than, raw TPS. A high-throughput chain that takes minutes to finalize a transaction can still leave your app in an uncertain state longer than a slower chain with fast finality.
There are two finality models in wide use. Probabilistic finality, used by Bitcoin and other proof-of-work chains, never mathematically guarantees a transaction can't be reversed. It just makes reversal exponentially harder with each additional block, which is why Bitcoin's 60-minute, six-confirmation convention exists as a practical threshold, not a hard rule. Deterministic finality, used by BFT-style protocols like Avalanche and Cosmos, gives a specific moment where the transaction is provably final, because a supermajority of validators has already signed off.
Ethereum normally reaches economic finality after roughly two epochs, commonly around 13 minutes, although the exact time depends on when the transaction enters the epoch cycle. Reversing a finalized Ethereum transaction would mean destroying a large fraction of staked ETH, which is what makes it economically, not just technically, irreversible.
Solana's current finality sits around 12.8 seconds. That's changing: the Alpenglow upgrade proposes a major redesign of Solana’s consensus and data-propagation architecture, including replacing TowerBFT with the Votor voting mechanism. After going live on a community test cluster in May 2026, Alpenglow feature-flag implementation remains under active testing on testnet, so sub-second finality should be treated as a target under development rather than a live mainnet metric.
The Decentralization Tradeoff
The blockchain trilemma says you can't maximize throughput, decentralization, and security all at once. What that looks like in practice is validator hardware requirements pushing out smaller operators as chains optimize for speed.
Solana requires enterprise-grade validator hardware, roughly 24+ core CPUs and 384GB+ of RAM according to current infrastructure guides, which is a meaningfully higher bar than running a home validator. That requirement, combined with the Solana Foundation pruning underperforming validators, is a documented factor in the network's validator count falling from over 2,500 in early 2023 to roughly 800 in early 2026, a decline of about 68%. The Nakamoto coefficient tracked that drop, falling from around 31 to around 20 over the same period.
Ethereum sits at the other end. Its hardware requirements are modest enough for home staking, which is part of why it supports over 1 million validator keys across roughly 10,000 physical nodes in more than 80 countries. That breadth is genuinely different from a single Nakamoto coefficient number, since it reflects validator-key concentration under a stake-weighted model rather than a simple entity count, and the two aren't directly comparable without saying so.
Cardano takes a third approach, using its "K" parameter as a direct economic lever: pools that grow past the saturation point start earning lower rewards per delegator, which pushes stake toward smaller pools. That mechanism is why Cardano can post a Nakamoto coefficient of 116, well above Solana's or Ethereum's raw validator-count comparisons, despite far lower raw throughput.
What The Nakamoto Coefficient Doesn't Tell You
The metric everyone quotes has real limits. A systematic review of blockchain trilemma measurement literature found that the Nakamoto coefficient only captures the minimum number of entities needed to compromise a network; it doesn't tell you how resources are distributed across everyone else, how geographically concentrated validators are, or whether one person or company quietly controls several of them.
A chain could have a high coefficient and still be more centralized in practice than the number suggests, if a handful of large operators run many nodes each. If you're doing real due diligence on a chain, the Nakamoto coefficient is a starting point, not the whole picture.
How To Verify These Numbers Yourself
None of the figures above should be taken on faith, and you don't need specialized tools to check them. For live TPS and finality, Chainspect's chain-comparison pages update from real block data. For Ethereum-specific throughput and gas usage, Etherscan's block explorer shows per-block transaction counts directly.
For governance changes like Cardano's K-parameter proposal, the chain's own governance forum and constitution repository are the primary source, not a summary article. We cover this verification process in more depth in our companion piece on auditing whitepaper and project claims against on-chain evidence.
Summary For Developers
- High-frequency trading or gaming: Solana or Sui offer the fastest real-time responsiveness and the lowest fees, at the cost of higher validator hardware requirements and a smaller validator set than Ethereum.
- High-value institutional settlement: Ethereum's slower finality is offset by deeper validation distribution and the largest economic security budget of any smart contract chain.
- Regulated or enterprise deployments: Avalanche's subnet model allows custom validator sets and compliance rules, which is part of why it's been used for institutional tokenized-fund pilots.
There's no universal "best" chain here. The right choice depends on whether your application cares more about raw speed, settlement certainty, or the ability to independently verify who's actually securing the network, and each of those tradeoffs is measurable, not a matter of marketing copy.
Frequently Asked Questions
Does higher TPS always mean lower transaction fees?
Not necessarily. Fees track demand relative to available block space, not raw throughput capacity. A high-TPS chain under heavy load can still have fee spikes if usage outpaces even its higher ceiling.
Why hasn't Bitcoin just increased its block size or speed to compete on TPS?
It's a deliberate tradeoff, not a technical limitation. Larger blocks or faster block times raise the hardware bar to run a full node, which is the exact centralization risk Bitcoin's design prioritizes avoiding over raw throughput.
Is a chain's finality time the same as its block time?
No, and conflating them is a common mistake. Block time is how often a new block is produced. Finality time is how long until that block becomes irreversible, which can be many multiples of the block time, as Ethereum's gap between 12-second blocks and 12.8-minute finality shows.
Related Articles
- Byzantine Fault Tolerance in Practice: Latency and Finality Tradeoffs Across PoW, PoS and BFT
- Smart Contract Architecture for Asset Custody: State Machines, Access Control and Upgrade Patterns
- Evaluating Cryptographic Claims in Whitepapers Against Verifiable On-Chain Evidence
- Oracle Design and Data Integrity: How Price and Asset Feeds Are Secured, and How They Fail
