Injective promises 25,000 transactions
a second. We measured 22

Injective's documentation says the chain sustains 25,000 transactions a second. Over 10,000 consecutive blocks this morning it settled 136,799 transactions in 6,091.5 seconds. That is 22.5 a second, a factor of about 1,113 out. Two independent public nodes returned the same heights, timestamps and counts, to the digit.

What the claim actually says

The figure is Injective's own. Its documentation lists “Throughput, sustained 25,000 tx/s” and “Throughput, theoretical upper bound 600,000 tx/s”, then adds: “The above stats are accurate as of April 2026. Note that they may change based on market movements and usage.” So the distinction between sustained and theoretical is drawn, and there is a staleness caveat. What is absent is any benchmark: no methodology, no hardware, no conditions, no report to read. The homepage, meanwhile, makes no throughput claim at all, and gives block time as 0.59 seconds where the documentation says 650 milliseconds and its own quickstart says 580.

The claim that survives

Block time does, comfortably. Across those 10,000 blocks they arrived every 0.609 seconds, quicker than the 650 milliseconds advertised, and the worst single gap was 2.31 seconds. Day to day, sub-second finality here is the measurement rather than the slogan — which makes it the more striking that the chain stopped dead for four hours last month.

What the traffic is

Throughput only matters if something uses it. Decoding 150 blocks inside the same window, and unwrapping the delegated-execution wrappers that hide much of the payload, gives 2,192 underlying messages from 2,038 transactions. Exchange orders and cancellations are 64.8 per cent of them, oracle price relays 26.5, Ethereum-compatible transactions 5.4 and CosmWasm calls 2.9. Ordinary person-to-person transfers did not appear at all.

And 68.4 per cent of transactions were signed by delegated authority: software, not somebody pressing a button. Injective calls itself a chain built for finance and the traffic agrees. This is an order book, busy for its size, and not a payments network.

The four hours that were not announced

On 31 August block production stopped twice: for 37 minutes and 46 seconds, then for 3 hours, 42 minutes and 12 seconds, both measured from the block timestamps. Secondary reports say Injective characterised this as an accelerated upgrade that overran; we could not retrieve the statement itself. Whatever it is called, the chain produced no blocks for about four hours and twenty minutes.

There is no incident history to consult: status.injective.network monitors endpoint availability, not the chain. The upgrade two days later promises “targeted chain improvements” and “enhancing protocol reliability”, and says nothing of an outage, though its release name is tagged safeharbor. Secondary reporting blames an exploit of permissionless binary-options markets costing roughly $4.9m, but those accounts disagree and no postmortem exists, so the cause is unestablished. Every such market is now marked demolished.

The economics are very small

Total value locked is $13.2m, 72nd of the 467 chains DefiLlama tracks. The whole application layer earned $186 of fees in a day and $11,601 over thirty days, against $40.7m in its lifetime: the revenue is historical. Helix, the flagship exchange, holds $784,000.

The supply nobody updated

CoinGecko gives INJ a total supply of 100,000,000, the genesis figure. Queried directly, on two nodes, the chain says 123,052,723. At £5.56 the real fully diluted value is about £684m rather than the £557m a price site displays, so anyone sizing this from a listing page is short by roughly £128m. Inflation runs at 4.40 per cent, or 5.41m INJ a year, while the burn auction retires around 328,500 — six per cent of issuance.

The verdict

Against it: a throughput figure wrong by three orders of magnitude, a four-hour halt with no postmortem, fees that round to nothing, a validator set capped at 45 where six entities could stop the chain, and 21 days to unstake during the next one.

For it: twelve mainnet upgrades in twelve months, each verifiable at the exact height governance approved, median downtime about two minutes. Ethereum and CosmWasm contracts both carrying live traffic. Two hundred and eighty-six active derivative markets. A block time better than advertised.

The engineering is real and the marketing is not. Injective is a fast, competent, small derivatives venue whose documentation describes a different chain. At £5.56 you are paying for the documentation.

Measured and retrieved 28 September 2026. Throughput and block time computed by us from Tendermint RPC over heights 184,911,045–184,921,044 at sentry.tm.injective.network, repeated on injective-rpc.publicnode.com with identical results. Message types decoded from the raw payloads of heights 184,920,895–184,921,044. Market counts, supply, validators and inflation from the chain's own modules on both nodes. Claims quoted from docs.injective.network and injective.com; upgrade text from governance proposal 690. The 31 August timestamps come from Injective's explorer API, because every public node we found prunes history short of that date — and that indexer is Injective's own. Price from CoinGecko; value locked from DefiLlama. Contested: the documentation gives block time as both 650 ms and 580 ms and the homepage as 0.59 s; DefiLlama's chain figure of $13.2m and its protocol entries' $33.6m count different things; CoinGecko's supply disagrees with the chain by 18.7 per cent and we have used the chain's. The halt durations are ours; its cause rests on secondary reports that conflict. Not a recommendation to buy, sell or hold anything.