Sign in
Article On-chain

Bitcoin RBF and CPFP explained: how to unstick a transaction

Bitcoin RBF and CPFP explained: how they differ, how to size the fee bump and what changed in Bitcoin Core. Live Nov 22, 2026 example: a 364-sat replacement.

Bitcoin RBF and CPFP explained: how to unstick a transaction
Contents
  1. Why a bitcoin transaction gets stuck
  2. How RBF works: replacing the transaction
  3. How CPFP works: child pays for parent
  4. RBF vs CPFP: which one to use
  5. How to speed up a stuck bitcoin transaction, step by step
  6. What it looks like today: a live mempool example
  7. Limitations and common mistakes
  8. FAQ

Bitcoin RBF and CPFP are the two standard ways to speed up a transaction that is stuck in the mempool because its fee is too low. RBF (replace-by-fee) swaps the transaction for a new version that pays more, and it is the sender's tool. CPFP (child pays for parent) leaves the original untouched and attaches a high-fee "child" that drags the "parent" into a block, so the recipient can use it too. As of Nov 22, 2026, 3:00 AM ET, the network was quiet: mempool.space recommended 1 sat/vB for every priority level, which put the cost of bumping an ordinary payment at a few cents. This guide covers how each method works, what it costs and which Bitcoin Core rules have changed over the past two years.

Why a bitcoin transaction gets stuck

Miners fill blocks with the transactions that pay the most per unit of block space, measured in satoshis per virtual byte (sat/vB). Everything else waits in the mempool; our guide to Bitcoin mempool fees explains that queue in detail. If your fee rate is lower than the rates ahead of you, your transaction sits there until the backlog clears.

The wait is long but not endless. By default a Bitcoin Core node drops a transaction from its mempool after 336 hours, or 14 days (the -mempoolexpiry setting). Your coins are not lost when that happens: an unconfirmed transaction changes nothing on the blockchain, and the same inputs can be spent again. Until the payment has confirmations, you can bump it in one of two ways.

How RBF works: replacing the transaction

RBF is a node policy that lets an unconfirmed transaction be replaced by a different one that spends at least one of the same inputs and pays a higher fee (Bitcoin Optech). In practice the wallet shrinks your change output and hands the difference to the miner. The recipient and the amount stay the same; the fee and the transaction ID (txid) change.

The current replacement conditions are listed in the Bitcoin Core policy docs. The ones that matter to a user:

  1. The replacement pays an absolute fee at least equal to the total paid by the transactions it evicts.
  2. The extra fee covers the replacement's own relay cost at the incremental relay feerate, 0.1 sat/vB by default. For a 500 vB transaction that is at least 50 satoshis on top of the old fee.
  3. The replacement conflicts with no more than 100 clusters of related transactions.
  4. The mempool must be strictly more profitable to mine afterwards. For a standalone transaction, a higher fee and a higher feerate than the original are enough (Bitcoin Core 31.0 release notes).

Most people bump by far more than the minimum: the goal is a rate that actually gets into the next few blocks.

The RBF flag is no longer required

Replacement used to be allowed only if the transaction signaled it. Under BIP 125, at least one input had to carry an nSequence value below 0xfffffffe. That was opt-in RBF. Then the defaults moved:

Bitcoin Core version Released What changed
28.0 Oct 2, 2024 Full RBF on by default: nodes replace any transaction, flagged or not
29.0 Apr 14, 2025 The -mempoolfullrbf option removed; full RBF is standard behavior
30.0 Oct 10, 2025 Minimum relay feerate and replacement increment cut to 0.1 sat/vB; bumpfee no longer needs a BIP 125 signal
31.0 Apr 19, 2026 Mempool rebuilt as "cluster mempool"; a replacement is accepted only if it improves the mempool as a whole

The flag itself is still everywhere. According to figures cited in the Bitcoin Optech newsletter of June 19, 2026, about 75% of transactions still set the signaling nSequence value. It just no longer decides whether a replacement is possible. What matters now is whether your wallet can build and broadcast one.

How CPFP works: child pays for parent

CPFP does not touch the stuck transaction. You spend one of its unconfirmed outputs in a new transaction with a high fee. Consensus rules say a child cannot be mined before its parent, so a miner who wants the child's generous fee has to include both. Nodes score them as a package by their combined feerate, which Bitcoin Optech calls ancestor feerate mining.

That is the key difference from RBF: anyone who controls an output of the stuck transaction can do CPFP. The recipient spends the incoming payment; the sender spends the change. If an exchange sent your withdrawal with a low fee, you cannot replace its transaction, but you can bump it with a child, provided the coins went to a wallet you control.

The child's fee is worked out like this:

child fee = target feerate × (parent size + child size) − parent fee

Example: the parent is 200 vB and pays 0.2 sat/vB, or 40 satoshis. You want a combined rate of 2 sat/vB, and a one-input, one-output child takes about 110 vB. The package needs 2 × (200 + 110) = 620 satoshis, 40 of which are already paid, so the child must pay 580 satoshis, roughly 5.3 sat/vB on its own. This is why CPFP always costs more than RBF for the same target: you are buying block space for two transactions, not one.

RBF vs CPFP: which one to use

RBF CPFP
Who can use it Sender only (needs the keys to the inputs) Sender (via change) or recipient
What happens Old transaction is replaced A child is attached to the old one
Does the payment txid change Yes No
Cost Top-up to the target rate Top-up plus space for a second transaction
Needs a spendable output No Yes, your own output in the stuck transaction
Best when You sent it and your wallet has a "bump fee" action You were paid with a low fee, or the txid must not change

The short rule: if you created the transaction, use RBF because it is cheaper. If you are the recipient, or the sender cannot replace it, CPFP is what is left.

How to speed up a stuck bitcoin transaction, step by step

  1. Look the transaction up by txid in a block explorer and compare its sat/vB rate with current network rates. Our dashboard shows network fees in the Numbers block.
  2. If the rate is close to what the next blocks are taking, wait. It may confirm on its own.
  3. If you are the sender, find the "Bump fee" or "Increase fee" action in your wallet. In Bitcoin Core it is the bumpfee command; Electrum and Sparrow have it in the transaction menu. Set a new rate and sign the replacement. With a hardware wallet or multisig, you sign the replacement the same way as a normal payment, for example with a PSBT.
  4. If you are the recipient and your wallet lets you spend unconfirmed coins, send the incoming payment to a fresh address of your own at a higher rate, using the formula above. Electrum and Sparrow offer a dedicated CPFP action.
  5. Check that the explorer shows the replacement or the child, then wait for a confirmation.

There is a third route: paid accelerators that arrange inclusion with mining pools directly. Mempool Accelerator takes payment off-chain, so no new transaction has to be signed. That means trusting a middleman and specific pools rather than relying on a protocol rule.

What it looks like today: a live mempool example

On Nov 22, 2026, at about 3:07 AM ET, the mempool.space mempool held roughly 62,000 transactions totaling almost 30 million vB, or about 30 blocks of backlog. Nearly all of it paid less than 1 sat/vB: the median rate in the projected next block was about 0.12 sat/vB. Rates that low became possible after Bitcoin Core 30.0 cut the minimum relay feerate to 0.1 sat/vB.

Here is a real replacement from the mempool.space RBF feed. At 3:00 AM ET a 372 vB transaction was broadcast with a 73-satoshi fee, under 0.2 sat/vB. After 171 seconds the sender replaced it with a new version of the same size paying 364 satoshis, or 0.98 sat/vB. The top-up was 291 satoshis against a required minimum of 37.2 (372 vB × 0.1 sat/vB). With bitcoin at $59,170 as of 3:00 AM ET, the whole replacement fee came to about $0.22 and the top-up to about $0.17.

On a quiet network a fee bump costs next to nothing. On a busy one the same logic runs with different numbers: with the queue at 50 sat/vB, replacing that same transaction would cost 18,600 satoshis. For the live price and network state, see the Bitcoin page.

Limitations and common mistakes

  • RBF is not a guaranteed cancel. "Cancelling" means replacing the payment with one that sends the coins back to yourself. It works only if no miner has confirmed the original first. After one confirmation, neither RBF nor CPFP applies.
  • An unconfirmed payment is not a payment yet. With full RBF any unconfirmed transaction can be replaced, so a merchant should wait for at least one block. For instant payments there is the Lightning Network.
  • A higher rate is not enough; the absolute fee must rise too. A smaller replacement that pays fewer satoshis in total is rejected.
  • CPFP needs a wallet that spends unconfirmed coins. Many wallets hide such outputs by default, and exchanges will not let you spend a deposit before it is credited.
  • Long chains hit limits. Since Bitcoin Core 31.0, a connected group of unconfirmed transactions (a cluster) is capped at 64 transactions and 101 kB of virtual size, and the old CPFP carve-out exception is gone.
  • Nodes differ. Replacement rules are node policy, not consensus. Some operators keep the old 1 sat/vB minimum, and a transaction below it may propagate less well.
  • Privacy. A replacement that adds a new input, or a child transaction, links your coins together in public.

Complex protocols such as Lightning, which must add fees to pre-signed transactions, rely on separate machinery: version 3 transactions (TRUC, BIP 431) and one-parent-one-child package relay, both introduced in Bitcoin Core 28.0. Ordinary users do not need them, but they are the reason a parent with a near-zero fee can reach a miner together with its child at all.

FAQ

Can I speed up a transaction if my wallet has no RBF support? Yes. Nodes running Bitcoin Core 28.0 or later accept a replacement for any transaction, so you can restore the wallet from its seed phrase in software that can bump fees, or use CPFP.

How many times can a transaction be replaced? There is no formal counter, but each replacement must pay more than the last by at least 0.1 sat/vB of its own size, so every further attempt costs more.

What happens to the old transaction after RBF? Nodes drop it from the mempool. Only one version can ever be mined because both spend the same inputs. The recipient should watch the address, not the old txid.

What if I do nothing? The transaction either confirms once the queue falls to its rate, or drops out of most nodes' mempools after 14 days. The coins are then spendable again, though your wallet may need a rescan.

This is not investment advice.

Morning brief and news on TelegramEvery morning at 9:00 AM ET, before the US open: price, 24h scenario, squeeze risk. Key news around the clock.
Coins
Category
Tags
#RBF#CPFP#fees#mempool#BTC
Ryan Mitchell
On-chain and mining

Breaks down blockchain and mining data: hashrate, difficulty, holder and whale behavior.

Comments

Read next