Bitcoin fee estimation API and transaction fee rate data

BitcoinDatabase indexes the confirmed chain and the unconfirmed set together, so the fee rate you quote a customer and the fee rate you later reconcile against come from the same place. GET /v1/mempool returns live mempool state with fee-rate estimates, GET /v1/tx/{txid} returns the fee and weight of any transaction, and the same data is queryable in SQL so you can audit what you actually paid over any window. Informational on-chain data only, not financial advice.

The reason this page exists is that the public fee estimators currently disagree with each other by a factor of four, and all of them sit well above what the network is charging. We read three of them within the same second on 17 September 2026, at block height 967,354, with all three providers agreeing on the tip. For the question "what gets into the next block", mempool.space returned 1 sat/vB, Blockstream Esplora returned 3.119 sat/vB, and BlockCypher returned 4,003 satoshis per kilobyte, which is 4.003 satoshis per byte. Same instant, same chain, same question, a 4.0x spread.

The spread is not the worst of it. BlockCypher’s cheapest tier, low_fee_per_kb at 1,348 sat/kB, is 35 percent more expensive than mempool.space’s most urgent tier. A team that picks the conservative provider and then picks its economy setting still pays more than a team that picks the aggressive provider and asks for the fastest confirmation available.

So we measured what the network actually charged. Across the fifteen consecutive blocks from 967,340 to 967,354, the median fee rate paid came in below Esplora’s next-block estimate in fifteen blocks out of fifteen, and below BlockCypher’s high tier in fifteen out of fifteen. The median of those fifteen block medians was 0.348 sat/vB. Against that, BlockCypher’s high tier was quoting 11.5 times the going rate and Esplora’s next-block figure was 9.0 times it.

Fourteen of those fifteen blocks confirmed at least one transaction at or below 1 sat/vB, and the cheapest confirmed transaction we recorded paid 0.307 sat/vB. That number matters because most fee guidance still in circulation, including several of the pages ranking today for this topic, states that nodes enforce a 1 sat/vB relay floor. They no longer do. Bitcoin Core pull request 33106 lowered the default minrelaytxfee and incrementalrelayfee to 0.1 sat/vB, a 90 percent cut, and Bitcoin Core 31.0, released 19 April 2026, went further and updated the fee estimator itself. Its release notes say the estimator minimum fee rate bucket "was updated from 1 sat/vB to 0.1 sat/vB, which matches the node’s default minrelaytxfee", and that estimates may now return sub-1 sat/vB values.

Bitcoin Core moved. The public REST estimators, on the evidence above, have not. mempool.space returned exactly 1 for all five of its tiers, fastest through minimum, on two separate reads minutes apart, which is an estimator pinned to a floor the protocol has already lowered rather than one tracking a market.

In money: a standard single-input, two-output native SegWit spend is about 141 vB. At the 0.348 sat/vB median actually paid, that is 49 satoshis, roughly 3.7 US cents with bitcoin at 76,488 dollars. At BlockCypher’s high tier it is 564 satoshis, about 43 cents. The gap is 515 satoshis per withdrawal. A desk pushing a thousand unbatched withdrawals a day is leaving roughly 0.005 BTC a day on the table, about 394 dollars, and a little over 143,000 dollars across a year. Batch your payouts and that shrinks a lot, which is itself an argument for measuring rather than trusting a tier label.

Two honest caveats. First, this is one reading of a quiet mempool, not a law: when real competition for block space returns, the conservative estimators will be closer to right and paying 0.3 sat/vB will strand your transaction for days. The point is not that fees are always cheap, it is that a hardcoded tier from a third party is not a substitute for reading the mempool you are actually broadcasting into. Second, a lower relay floor only helps if the nodes between you and a miner also run it. A 0.3 sat/vB transaction that will not propagate is not cheap, it is invisible.

We re-measured the next day, as that first caveat demands. At 13:03 UTC on 18 September 2026, at block 967,554 with all three providers again on the same tip, mempool.space quoted 5 sat/vB for its fastest tier, BlockCypher 2,856 sat/kB (2.856 per byte) and Esplora 1.069. The median fee rate paid across the thirty blocks 967,525 to 967,554 had risen to 1.02 sat/vB. Esplora was now close to the market, BlockCypher 2.8 times above it, and mempool.space, the lowest quote the day before, was the highest at 4.9 times. The ranking of the estimators flipped inside 24 hours. That is the strongest argument on this page for auditing your realized fee rate against confirmed blocks rather than choosing a provider and trusting it.