Since 5 November 2026, twelve confirmations on Base take 2.4 seconds. Before the Denim upgrade they took 24 seconds, and neither figure tells you whether a deposit is final: Base's own finalized head was running 15 to 18 minutes behind the tip when I measured it this morning.
Block counts are how most deposit pipelines decide when to credit a customer, and three upgrades in the last nine weeks have quietly changed what those counts mean. On 22 November 2026, between 08:11 and 08:14 UTC, I polled public RPC endpoints on eight chains six times, 30 seconds apart, and compared the latest block with each chain's safe and finalized heads. I then lined the results up against the deposit rules Kraken, Circle Mint, Binance and Coinbase publish. The pages ranking for "how many confirmations USDC" give a single number per chain. None of the ones I read showed the code that checks a block is final, and one still says Base deposits take seven days.
How Many Confirmations Does a USDC Deposit Need?
Stop counting blocks and ask the chain for its finality signal. Here is what that signal looked like this morning.
| Chain | Avg block time (measured) | Finality signal | Tip to final, measured |
|---|---|---|---|
| Ethereum | 12.0 s | finalized block tag | 70 to 83 blocks, 14.0 to 16.6 min |
| Base | 0.2 s | finalized (follows L1) | 4,630 to 5,409 blocks, 15.4 to 18.0 min |
| Arbitrum One | 0.3 s | finalized (follows L1) | 3,436 to 3,952 blocks, 17.2 to 19.7 min |
| Polygon PoS | 1.5 s | finalized (Heimdall milestones) | 1 to 2 blocks, 1 to 3 s |
| Avalanche C-Chain | 1.0 s | block acceptance | about 1 s, per Avalanche's docs |
| Solana | ~0.2 s slots | finalized commitment | 1 to 2 slots |
| Tron | 3.0 s | /walletsolidity/* endpoints | 18 to 19 blocks, 54 to 57 s |
Stellar is the eighth: ledgers close every 5 seconds and a closed ledger is final, because its consensus protocol halts rather than forks.
The spread is the point. "Final" ranges from about a second on Polygon and Solana to around 20 minutes on an optimistic rollup. A single "12 confirmations" policy would be about 2.4 minutes on Ethereum and 2.4 seconds on Base.
Confirmations vs Finality: Why Block Counts Broke in 2026
Confirmation counts come from proof-of-work, where every extra block made a reorg exponentially less likely and there was no other signal to use. Most chains a stablecoin desk touches now have explicit finality, and the count is a proxy for something you can query directly.
The proxy is also getting worse, because block cadence keeps changing underneath it:
- Base Denim (5 November 2026) moved Base to five canonical blocks per second and retired Flashblocks. Its preconfirmation WebSocket endpoints shut down on 19 November, and the
pendingtag now resolves to the latest canonical block. - Solana Alpenglow (live on mainnet-beta since 21 October 2026) replaced Tower BFT. Anza's docs now state that the confirmed and finalized commitment levels are equivalent.
- Avalanche Helicon (22 September 2026) split execution from acceptance on the C-Chain.
safeandfinalizednow point at the last settled block, a few blocks behindlatest, and Avalanche's docs say plainly that these tags do not mark finality.
Ethereum Finality: safe vs finalized Block Tags
Ethereum finalises at epoch boundaries. An epoch is 32 slots of 12 seconds, and a checkpoint is justified, then finalized, by successive votes from validators holding two-thirds of staked ETH. Reverting a finalized block would cost at least a third of all staked ETH. ethereum.org quotes "about 15 minutes", and my samples ran from 14.0 to 16.6 minutes.
The JSON-RPC spec gives you three useful tags for eth_getBlockByNumber:
latest: the head. It can reorg.safe: the justified checkpoint, safe "under honest majority and certain synchronicity assumptions". The consensus spec calls it a stopgap. It ran 7.6 to 10.2 minutes behind the head this morning.finalized: "cannot be re-orged outside of manual intervention".
import { createPublicClient, http } from "viem";
import { mainnet } from "viem/chains";
const client = createPublicClient({ chain: mainnet, transport: http(RPC_URL) });
async function isFinal(txHash: 0x${string}): Promise {
const receipt = await client.getTransactionReceipt({ hash: txHash });
if (receipt.status !== "success") return false;
const finalized = await client.getBlock({ blockTag: "finalized" });
if (finalized.number < receipt.blockNumber) return false;
const canonical = await client.getBlock({ blockNumber: receipt.blockNumber });
return canonical.hash === receipt.blockHash;
}
The failure mode to plan for is that finality can stop. On 12 May 2023, between 17:20 and 18:24 UTC, participation fell to 30.7% and the finalized checkpoint froze for about 64 minutes. A pipeline keyed on finalized stalls rather than crediting anything wrongly, which is the right failure, but you need an alert on finalized-head lag. I'd page at 25 minutes, roughly four epochs, which is about where the protocol's inactivity leak starts.
Is a Base or Arbitrum Deposit Final Before It Reaches Ethereum?
No. Until its batch is posted to Ethereum and that Ethereum block finalises, an L2 block rests on the sequencer's word. The seven-day window people quote applies to withdrawals through the canonical bridge, not deposits. Optimism's docs say OP Stack transactions are not reorganised in response to fault-proof challenges, and Arbitrum's say the same about fraud proofs.
Both stacks expose the same three tags, but safe does not mean the same thing on each, and my samples showed it:
| L2 | safe means | Measured safe lag | Measured finalized lag |
|---|---|---|---|
| Base (OP Stack) | batch included in an L1 block | 18 to 62 s | 15.4 to 18.0 min |
| Arbitrum One | batch posted and that L1 block is L1-safe | 10.2 to 12.8 min | 17.2 to 19.7 min |
OP Mainnet, on the same stack as Base, showed a finalized lag of 17.7 to 21.0 minutes. If your code treats safe as one risk tier across both chains, Base deposits will clear about ten minutes earlier than Arbitrum deposits, at a weaker guarantee. Pick finalized for anything a customer can withdraw, or document why you accept less.
Solana Commitment Levels After Alpenglow: confirmed vs finalized
Under Tower BFT, confirmed meant a supermajority had voted on the block and finalized meant 31 or more confirmed blocks had been built on top of it, roughly 13 seconds later. Most guides still describe that model.
Under Alpenglow, a block finalises when 60% of stake votes to finalise it, or 80% votes to notarise it. In every sample this morning, getSlot with confirmed and finalized returned the same slot, one or two behind processed. getRecentPerformanceSamples also showed numTransactions equal to numNonVoteTransactions: vote transactions have left the ledger. The node reported solana-core 4.4.0 and produced about 4.6 slots per second.
For deposits, poll getSignatureStatuses and credit when confirmationStatus is "finalized". You now lose almost nothing by waiting for it. The Rust RpcClient::new already defaults to finalized, but check your TypeScript client's default before you assume the same.
Tron USDT Confirmations: Solidified Blocks and /walletsolidity
Tron has 27 Super Representatives producing 3-second blocks. A block becomes irreversible once 70% of active SRs, 19 of 27, have produced blocks at or above its height. I measured a constant 19-block gap, 57 seconds, in five of six samples.
Tron exposes finality as a separate API surface rather than a tag. Every /wallet/ HTTP call has a /walletsolidity/ twin that only sees solidified blocks:
curl -s -X POST https://api.trongrid.io/walletsolidity/gettransactioninfobyid \
-d '{"value":""}'
Two gotchas. An empty {} response means the transaction is not solidified yet, so do not mark the payment failed. And Tron's JSON-RPC accepts finalized on eth_getBlockByNumber but only latest on eth_getBalance and eth_call, so read solidified balances from /walletsolidity/getaccount. The USDT TRC20 fee guide covers the other half of Tron's quirks.
Polygon, Avalanche and Stellar: Fast Finality, With Caveats
Polygon PoS now finalises through Heimdall milestones, which need two-thirds of validators and, per Polygon's docs, typically land in 2 to 5 seconds. The finalized head was 1 to 2 blocks behind the tip in my samples. Polygon earned its conservative reputation: a 157-block reorg at height 39,599,624 in February 2023. But that predates the July 2025 Heimdall v2 upgrade, which cut finality from 1 to 2 minutes to about 5 seconds. Avalanche C-Chain is final in about a second, when Snowman accepts the block, and Snowman does not reorg. After Helicon, thefinalized tag sat 5 to 8 blocks behind latest, so polling it adds around 6 seconds and no safety.
Stellar cut mainnet ledger close from about 5.7 seconds to 5 seconds with Protocol 28 on 16 September 2026. One ledger is final.
Kraken vs Circle vs Binance: Deposit Confirmations Compared
This is what each provider published on 22 November 2026. Binance's figures come from the public configuration behind its deposit page: the first number credits the deposit, the second releases it for withdrawal.
| Network | Kraken | Circle Mint | Binance (credit / release) |
|---|---|---|---|
| Ethereum | finalization, 64 to 96 blocks, ~20 min | 12 (~3 min) | 6 / 64 |
| Base | "finalized", ~20 min | 12 ETH blocks after posting | USDC 1 / 100 |
| Arbitrum One | "finalized", ~20 min | 12 ETH blocks after posting | 120 / 120 |
| Polygon PoS | 250 (~25 min) | 2 to 3 (~8 s) | 200 / 300 |
| Solana | under 1 min | 1 (~200 ms) | 1 / 0 |
| Tron (USDT) | 20 (~1 min) | not listed | 1 / 0 |
| Avalanche C-Chain | 20 (~4 min) | 1 (~2 s) | 12 / 0 |
| Stellar | under 1 min | 1 (~5 s) | USDC 1 / 0 |
Coinbase's glossary gives one relevant figure, 14 confirmations for ETH and ERC-20 tokens. Stripe's stablecoin docs publish none.
The disagreements are bigger than they look. On Polygon, Kraken waits 250 blocks for a chain whose finality gadget answers in under 5 seconds, while Circle waits two or three. On Ethereum, Circle credits after 12 blocks, about 3 minutes by its own estimate and well before Casper finality, and accepts the small reorg risk in that window. Binance's Arbitrum threshold, 120 blocks at 0.3 seconds, is about 36 seconds of sequencer confirmations, well short of the 10-plus minutes Arbitrum's own safe tag took this morning. And if Binance set its 100-block Base release threshold before Denim, that window shrank from 200 seconds to 20 on 5 November without anyone editing it.
Circle has no Tron row because it stopped supporting USDC on Tron in 2024. On Tron, USDT is where the volume is.
When Should You Credit a Stablecoin Deposit? My Position
My view: credit on a cheap signal, release on finality, and write every threshold as a finality condition rather than a block count.
Binance's credit/release split is the right shape. Showing the customer their money after one Base block or six Ethereum blocks costs you nothing if they cannot move it until the chain says it is final. Your exposure is then bounded by the withdrawal path, not by reorg odds you estimated once and never revisited. Kraken points the same way. When I compared these tables for the stablecoin dev tools guide in September, it listed 30 blocks for Ethereum and 40 L1 blocks for Base. Both rows are now finality conditions.
My prediction: by the end of 2027, the major exchanges and issuers will express most rows in their deposit tables as finality conditions, not block counts. Three chains changed their block cadence or finality semantics in nine weeks this autumn, and a block count has to be re-derived every time one does.
What This Means for Payment Engineers
1. Store a finality rule per chain, not a count. For example: { chain: "base", credit: "latest", release: "finalized" }. Re-check the rules against release notes for every network upgrade.
2. Query finality directly. finalized on Ethereum, the OP Stack, Arbitrum and Polygon. confirmationStatus: "finalized" on Solana. /walletsolidity/* on Tron. Block acceptance on Avalanche, since its tag now lags for no benefit.
3. Re-verify the block hash at the deposit's height once it is final, and key attribution on (chain, tx_hash, log_index) as in the USDC payment matching guide.
4. Alert on finality lag. Ethereum above 25 minutes, any L2 above 40, Tron's solidified head more than 30 blocks behind.
5. Treat safe as chain-specific. It is a justified checkpoint on Ethereum, L1 inclusion on Base and L1-safe on Arbitrum.
6. Audit inherited thresholds on Solana, Base and Avalanche set before this autumn's upgrades. Most of them now mean something different.
I'm Tom Wang, and I write about the engineering underneath payment rails. Chains change their block times by governance vote, so re-measure the lag yourself before trusting any number here, including mine.