Hedera's homepage carries two numbers two tiles apart. One says “10,000+” transactions per second. The other says “351,653” transactions in the last 24 hours. Divide the second by 86,400 and you get 4.07 a second.
We measured it. Forty-five thousand consecutive record files from the public mirror node, spanning 27.1 hours to 07:03 UTC today, carried 392,281 transactions. That is 4.0196 a second, at a mean block time of 2.1687 seconds. A separate 15,000-block window gave 4.1959. Hedera's own counter and our measurement agree to within 1.3 per cent, and both sit about 2,500 times below the advertised figure.
The 10,000 is a throttle setting
The documentation does not pretend otherwise, which is the interesting part. Hedera's mainnet page says transactions are “currently throttled” and publishes the table: 10,000 a second for plain transfers and “other” crypto and consensus transactions, far less for the rest. Account creation is throttled to two a second, topic creation to five, file transactions to ten, minting a non-fungible token to fifty.
So a network advertising ten thousand a second publishes a standard account-creation ceiling of two — a 5,000-fold gap inside its own docs. In fairness the same page adds a parallel “high-volume” bucket whose capacity is “set by governance” and so not published at all, making two a second the only disclosed figure rather than a hard limit. The getting-started page hedges the headline to “~10,000 TPS”; marketing turns the tilde into a plus sign.
Hedera's own benchmark slide reaches ten thousand. Its footer reads: “Throttles set to 10,400 transactions/second”.
That slide, in the version 0.74 release notes, shows five bars headed “>10,000 txn/sec” and names a 33-node configuration. It measures the throttle, on a test network, under synthetic load. Nothing on the homepage says so.
About half the traffic is Hedera talking to itself
We sampled transaction records stratified across anchor points spread over the window, to avoid a time-of-day artefact. Two payer accounts dominate, 0.0.3229 and 0.0.256009, and every one of 200 sampled transactions from each carries a public memo naming a Kubernetes pod of Hedera's own mirror-node monitoring service. Three separate samples put their combined share at 46.1, 50.0 and 53.5 per cent — so about half of what Hedera counts is Hedera pinging itself, a share that moves with the window and that we would not quote to a decimal place.
Of 1,500 recent transfers, the median moved one tinybar — a hundred-millionth of an HBAR — while paying roughly 96,700 tinybars in fees. Eighty-nine per cent moved a millionth of an HBAR or less.
What is genuinely good
The engineering is not the problem. Across 45,000 blocks the longest gap between them was 5.086 seconds, with none beyond six and no empty blocks — and that held while an IBM-operated consensus node sat in partial outage on Hedera's status page for about 29 hours. Burst capacity is real too: the busiest record file carried 2,551 transactions in 1.962 seconds. The last mainnet incident was 45 minutes in May.
The economics are thinner. DefiLlama puts chain value locked at $41.4m, 47th of 468 chains; SaucerSwap is 99.4 per cent of daily exchange volume and Circle's USDC 99.66 per cent of the $47.7m of stablecoins. DefiLlama puts the base layer's own fees at about $600 a day, against a $4.56bn market capitalisation. HBAR traded at £0.078340 at 07:02 UTC.
Thirty nodes, and none of them yours
There are 30 consensus nodes, every one run by a Hedera Council member, and a 450m HBAR cap flattens stake deliberately, so the Nakamoto coefficient is nine. It hardly matters. Hedera's operator documentation says mainnet “is currently comprised of permissioned consensus nodes operated by the Hedera Council”. Its governance FAQ says “initially, only Council members will run network nodes”, gives no date, and still refers to a “mainnet beta period”.
One last figure. Every number above comes from infrastructure Hedera owns, because the two public mirror endpoints resolve to one IP address and every third-party mirror wants a paid key. For a network sold on enterprise trust, that is the least comfortable measurement of all.
| Advertised throughput | 10,000+ a second |
| Measured, 45,000 blocks over 27.1 hours | 4.0196 a second |
| Hedera's own 24-hour counter | 351,653 = 4.07 a second |
| Documented account-creation throttle | 2 a second |
| Mean block time, longest gap | 2.1687 s, 5.086 s |
| Traffic from Hedera's own monitor | 46–53% across three samples |
| Median value moved per transfer | 1 tinybar |
| Consensus nodes, all Council-operated | 30 |
| Chain value locked | $41.4m |
| Base-layer fees, DefiLlama | about $600 a day |
| HBAR price | £0.078340 / $0.103561 |
Retrieved 5 October 2026. Throughput, block times, transaction types, payer concentration, supply and nodes were measured from the Hedera public mirror node REST API; the 45,000-block window ran 2026-10-04T03:57:11Z to 2026-10-05T07:03:42Z. Claims quoted from hedera.com and docs.hedera.com, including the mainnet throttle table and the version 0.74 performance slide. Price from CoinGecko; value locked, volume, stablecoins and base-layer fees from DefiLlama; node status from status.hedera.com. The monitoring share is a range because three stratified samples of the same window returned 46.1, 50.0 and 53.5 per cent; it is about half, and we claim no more precision. Unverified: no independent cross-check was possible, because both public mirror endpoints resolve to one IP address and every third-party mirror needs a paid key, so every chain figure derives from Hedera-operated infrastructure. The homepage's “71+ billion” cumulative transactions is unverified and not relied on. Mainnet self-reports version 0.77.5 while the release notes document only 0.77.2. Not a recommendation to buy, sell or hold anything.