Base shipped three of its own hardforks in 2026: Azul on May 28, Beryl on June 25 and Cobalt on September 30. Together they shortened withdrawals to Ethereum, added a native token standard and introduced conditional transactions. The next one, Denim, would make Base produce a full block every 200 milliseconds. As of October 9, 2026, Base is targeting November 2026 for mainnet.

Last updated: October 9, 2026. Dates come from Base's own upgrade docs and node release notes.

Key takeaways

  • Base decoupled from Optimism's Superchain governance in March 2026 and now names its own upgrades. Azul was the first.
  • Since Beryl, a standard withdrawal to Ethereum finalizes after 5 days with one proof, or 1 day when Base's TEE and ZK proofs both back the same state. Before Azul it was 7 days.
  • Cobalt added validity transactions: signed transactions that Base considers for inclusion only while on-chain conditions match.
  • Denim is planned, not live. It replaces 2-second blocks and Flashblocks with five full blocks per second, so anything that counts time in blocks will run 10x faster.
  • L2BEAT still rates Base Stage 1. It runs a single sequencer, and its contracts can be upgraded with no exit window.

Base in 2026 at a glance

Upgrade Base Sepolia Base Mainnet Headline change
Azul April 20, 2026 May 28, 2026 TEE + ZK multiproofs, per-transaction gas cap
Beryl June 18, 2026 June 25, 2026 B20 tokens, 5-day single-proof withdrawals, Reth V2
Cobalt September 23, 2026 September 30, 2026 Validity transactions, B20 improvements
Denim Planned: October 2026 Planned: November 2026 200ms canonical blocks, Flashblocks removed

Denim's dates are planning targets. On October 1, Base moved the mainnet target from October to November. As of October 9, it has not published activation times or required client versions.

Update log: October 9, 2026: first published, covering Cobalt on mainnet and Denim's November target.

Why Base now names its own upgrades

Until 2026, Base adopted Optimism's OP Stack hardforks, such as Ecotone, Fjord, Isthmus and, in December 2025, Jovian. L2BEAT dates "Base leaves the Superchain" to March 4, 2026, describing it as Base decoupling from Superchain governance "with its own upgrade path."

Base's docs call Azul its "first independent network upgrade." Only Base's own node clients, base-reth-node and base-consensus, support it, so operators running op-geth, op-reth, Nethermind or other clients had to migrate.

Azul (May 28): two proofs and a gas cap

Azul changed how Base proves its state to Ethereum. A contract called the AggregateVerifier can check one or two proofs for the same state. A TEE prover running in AWS Nitro Enclaves handles the normal path. A permissionless ZK prover (L2BEAT identifies it as SP1) is the backstop that can dispute a bad claim.

When both proofs back the same state, a withdrawal to Ethereum can finalize after 1 day. With one proof, Azul kept the 7-day window, which Beryl later cut to 5. There is a catch: Base's proposer currently opens every proposal with a TEE proof only. A ZK proof can be added later, and Base says it may build ZK proofs into new proposals in future so that all of them get the faster path. Deposits and ordinary Base transactions never had this wait.

Azul also:

  • Lowered the per-transaction gas cap to 16,777,216 gas (2^24, EIP-7825), matching Ethereum. Transactions above it are rejected (deposits are exempt), so very large single transactions, such as big batch airdrops or smart-wallet bundles, must be split.
  • Doubled the gas cost of the secp256r1 precompile, which verifies passkey signatures, from 3,450 to 6,900 to match Ethereum's pricing.

An honest note on decentralization. Faster withdrawals do not make Base trustless. L2BEAT still puts Base at Stage 1, not Stage 2. Contract upgrades need sign-off from both Coinbase's Base Coordinator Multisig and the Base Security Council, and they take effect with no exit window. L2BEAT also notes that the Coordinator Multisig alone controls the allowlist of TEE provers. Base's docs say it runs a single active sequencer. If the sequencer censors you or halts, you can force a transaction in through Ethereum, with a maximum delay of about 12 hours.

Beryl (June 25): B20 and 5-day withdrawals

Beryl brought three changes:

  • B20, Base's native token standard. B20 tokens are ERC-20 compatible, but they run as Rust precompiles inside the node instead of as deployed Solidity contracts. They add a built-in compliance toolkit: transfer policies, freeze-and-reissue controls, role-based access, memos and supply caps. We cover it in B20 vs ERC-20: Base's new token standard.
  • Shorter single-proof withdrawals, down from 7 days to 5. The two-proof path stays at 1 day. Base says the change frees capital for fast-bridge liquidity providers.
  • Reth V2 as the reference execution client.

B20 did not switch on for mainnet on Beryl day. Base had planned activation for June 26, postponed it that day, then scheduled the mainnet Activation Registry for July 8, 2026 at 18:00 UTC. As of October 2026, Base's B20 docs list Mainnet among the networks where B20 is active.

A stability incident at launch, and what it teaches

On June 26, 2026, the day after Beryl went live on mainnet, Base delayed B20's mainnet activation, citing "an unrelated stability incident." Base's reliability docs link a postmortem of a "June 25th block production outage" on its engineering blog.

We could not open that postmortem, so we are not repeating press figures for how long blocks stopped or why. Base's own write-up is the place to read them.

The practical lesson for anyone launching a token or running a campaign: stay away from hardfork activation windows. Base announces timing in advance: Azul activated at 18:00 UTC on May 28, and node release notes told operators to upgrade by 18:00 UTC on June 25 for Beryl and by 18:00 UTC on September 30 for Cobalt. Check status.base.org, which reports block production, deposits and withdrawals, before anything time-sensitive.

Cobalt (September 30): "only if" transactions

Cobalt's headline feature is validity transactions. You sign a transaction and attach conditions, and Base considers it for inclusion only when every condition matches on-chain state. The supported checks are balance, storage, block number and Flashblock index. Base suggests them for conditional swaps, withdrawals and other state-dependent actions.

Two limits matter:

  • Every validity transaction must expire within 60 seconds of submission. These are short-lived conditional orders, not week-long limit orders.
  • Matching conditions does not reserve block space or guarantee inclusion. A transaction that never lands pays no execution fee. One that lands and reverts pays as normal.

Cobalt also:

  • Extended B20 with scheduled multiplier updates using the ERC-8056 interface, a new seize function (seizeWithMemo) that supersedes the now-deprecated burnBlocked, and UNION and INTERSECT policies that combine lists, such as "on the KYC list and not on the sanctions list."
  • Moved TEE signer registration to fully on-chain verification of AWS Nitro attestations, removing a dependency on an off-chain proving service.
  • Added dynamic upgrades, where nodes read upgrade timestamps from an Ethereum contract. On mainnet this runs in metrics-only mode for now.

Denim (planned): 200ms blocks and the end of Flashblocks

Today Base produces one canonical block every 2 seconds, with 10 Flashblocks inside it giving apps preconfirmations roughly every 200ms. Denim replaces that with five complete canonical blocks per second, each with its own hash, state root and receipts. Flashblocks streams and the "pending" preconfirmation state go away.

Denim also adds a BaseTime predeploy so contracts can read millisecond time, plus millisecond timestamp fields in RPC responses. The EVM's block.timestamp still counts whole seconds. On the token side, Denim tightens B20: transfers, mints and seizes to a token's own address will revert, the transfer-executor policy will apply on every transfer path, and any policy can be inverted with a NOT flag.

Denim is not live. As of October 9, 2026, Base lists Sepolia as planned for October and mainnet for November, and says Denim is not active on either network. For now, 200ms blocks run on Vibenet, Base's experimental preview devnet (chain ID 84538453). Vibenet also runs EIP-8130, an experimental form of native account abstraction that lets smart accounts work without bundlers or relays. EIP-8130 is not listed as part of Denim.

Who must act before Denim. Base warns that block.number will advance 10x faster, so any duration measured in blocks shrinks to one tenth:

  • Timelocks, cooldowns and deadlines written as block counts
  • Vesting or emission schedules measured in blocks
  • Per-block rate limits, whose effective per-second limit rises tenfold
  • Governance voting windows, including ERC20Votes with its default block-number clock

Logic based on timestamps (block.timestamp) is unaffected. Block counts in immutable contracts cannot be retuned after activation, so check them now. Bots, wallets and indexers that use the "pending" tag or subscribe to Flashblocks need to move to canonical data, such as newHeads and logs subscriptions.

Fees and capacity today

As of October 9, 2026, Base's docs list:

  • about 400M gas per full block
  • 2-second blocks with 10 Flashblocks each, and about 200ms preconfirmation latency
  • a reorg rate under 0.1%
  • multiple sustained bursts above 5,000 TPS

Base also put a floor under fees. It enabled a minimum base fee on December 4, 2025 and raised it in steps to 5,000,000 wei (0.005 gwei) on February 19, 2026, so the base fee no longer drops to extremely low levels when the chain is quiet. The floor is still small: at that rate, one million gas costs 0.000005 ETH in base fee, before the Ethereum data fee and any tip.

Beyond the protocol

Two non-protocol changes, as recorded in Base's own repositories:

  • Base Account is now Coinbase Wallet. The Base Account SDK README says so. The npm package is still @base-org/account, and existing integrations do not need to migrate. On September 11, Base moved its wallet docs to Coinbase Developer Platform, where the Base Account SDK docs continue as the Coinbase Wallet SDK and Base MCP became Wallet MCP.
  • Mini apps are now standard web apps. Base's migration guide said that after April 9, 2026, the Base App treats every app as a standard web app regardless of Farcaster manifests. Metadata moved to Base.dev projects, notifications go to wallet addresses, and sign-in uses wagmi, viem and Sign-In with Ethereum.

The network itself is still called Base.

What it means for you

  • Holders: Base's upgrade notes list no action for ordinary holders. Treat any message telling you to "migrate" or swap tokens for an upgrade as a likely scam.
  • Traders: expect faster canonical blocks after Denim, and short-lived conditional orders through validity transactions where your app supports them.
  • Token creators: do not launch during an activation window. Before Denim, check any vesting, lock or staking contract tied to your token for block-number timers.
  • Builders: migrate off Flashblocks and the "pending" tag, replace block counts with timestamps where you can, and test on Vibenet.

A plain token is the easy case. A Base token made with Solbase is a fixed-supply ERC-20 with 18 decimals, no owner, no mint function and no block-based timers, so Denim does not change how it behaves. A launch costs network gas plus a flat 0.001 ETH platform fee, as the Solbase docs explain. Solbase is non-custodial and does not provide liquidity, list or promote tokens. Any vesting or lock contract you add separately is still worth checking. If you also follow Solana, our guide to Solana's 2026 upgrades covers the other side.

FAQ

How long does a Base withdrawal take now?

As of October 9, 2026, a standard withdrawal to Ethereum waits 5 days when one proof backs the relevant state, or 1 day when both TEE and ZK proofs do. Base's proposer currently starts each proposal with a TEE proof only, so plan for 5 days unless your bridge shows otherwise. Getting Base's state onto Ethereum usually adds about 20 to 60 minutes. Third-party fast bridges pay out sooner but add their own fees and trust assumptions.

Is Base decentralized?

Partly. L2BEAT rates Base Stage 1: permissionless proofs and a Security Council, but one sequencer and no exit window before upgrades. If the sequencer censors you, you can force a transaction in through Ethereum within about 12 hours.

When is Denim on mainnet?

Base's target is November 2026, with Base Sepolia in October. Both are planning targets with no published activation time as of October 9, 2026, and the mainnet target has already moved once.

Will Denim affect my token?

A standard ERC-20's balances and transfers work the same. What can change is any contract around the token that measures time in blocks, such as vesting, timelocks, staking cooldowns or ERC20Votes windows. Those would run about 10x faster unless updated.

What happened to Base Account and Base mini apps?

Base Account, the smart wallet developers integrate, is now branded Coinbase Wallet, with docs on Coinbase Developer Platform since September 11. Since April 9, the Base App treats all apps as standard web apps, so Farcaster manifests no longer control how they appear.

Sources