Glossary · Nodes and clients

Light client

Published 22 May 2025 · Updated 2 Nov 2026 · 14 min read

A light client is software that follows a blockchain without downloading and executing every block. It accepts headers through Bitcoin proof of work or Ethereum sync committee signatures. It checks requested data with Merkle proofs against those headers, while trusting the network's majority for transaction validity.

  • A Merkle proof shows inclusion, not transaction validity or whether data was left out.
  • An Ethereum light client trusts 2/3 of a 512-member sync committee and needs a recent checkpoint.
  • On 2 November 2026, ethereum.org calls none of the four light clients it lists production-ready.

What a light client checks

A light client keeps small block summaries, requests data with proofs from a full node, then checks those proofs against the summaries. Different chains deliver summaries differently. The technical terms are block header, Merkle root and Merkle branch. A light node runs this blockchain client software. Bitcoin calls the approach Simplified Payment Verification, or SPV.

Full nodeLight clientWallet that trusts an RPC endpoint
DownloadsFull nodeBlocks and state needed for independent validationLight clientHeaders or committee updates, plus requested dataWallet that trusts an RPC endpointResponses supplied by its endpoint
Checks consensus rulesFull nodeEnforces consensus rules and validates transactionsLight clientChecks consensus evidence, without executing transactionsWallet that trusts an RPC endpointDoes not independently check consensus rules
Checks that data is in the chainFull nodeChecks transactions within downloaded blocksLight clientChecks Merkle branches against accepted rootsWallet that trusts an RPC endpointAccepts the endpoint's claims without verification
TrustsFull nodeIts rules and possibly a starting checkpointLight clientConsensus majority and any required starting checkpointWallet that trusts an RPC endpointThe RPC provider for every chain claim
PrivacyFull nodeCan read locally rather than querying a serverLight clientDepends on filters, peers and requested dataWallet that trusts an RPC endpointExposes requests and network activity to its provider

The anchor depends on the chain. Bitcoin checks proof of work and follows the most-work chain. Ethereum requires BLS signatures from 342 of 512 sync committee seats to advance finalized headers. CometBFT checks signatures representing more than 2/3 of voting power. Mina and ZK light clients use succinct cryptographic proofs, or SNARKs. A Merkle branch links data to an already accepted header.

These checks do not validate the transactions themselves. An RPC-only wallet performs none of these checks. A server can omit activity, and Electrum trusts its server about unconfirmed transactions. A full node instead enforces the rules governing blocks and their contents. The ethereum.org nodes page says light nodes have the same security guarantees as a full node. On 2 November 2026, its dedicated guidance describes equivalence as a future goal, not a present guarantee.

How Bitcoin SPV works

Section 8 of the October 2008 white paper describes SPV for the Bitcoin network, without running a full node. Each block header occupies 80 bytes, including a 32-byte Merkle root. The paper estimates 4.2 MB of headers per year. On 2 November 2026, the complete header chain at height 974,285 occupies 77.94 MB.

Figure 1 · mempool.space data, checked by hashingClimb a Bitcoin Merkle treeTrace the transaction at position 1,500 in block 974,285 one level at a time or all at once, then toggle a changed byte.
Test
Block 974,285, transaction at position 1,500Not startedTransaction ID3937ffa9c91e451d38ca8ab42118f542856ec3969cfd614daa84a0f47aaafca81Sibling on the right56223c1ca0f1783a…732b573ee5adParent hashNot computed yet2Sibling on the right12ff6ad7e0120aa9…09aae53dc25cParent hashNot computed yet3Sibling on the leftd0985400b8d62089…ea1a26253853Parent hashNot computed yet4Sibling on the left6de5c39bdfe0b380…c32fb26bc143Parent hashNot computed yet5Sibling on the left72da2d1cf93b93a8…4e076de42168Parent hashNot computed yet6Sibling on the right441922d2666f492b…8171a6fa21fdParent hashNot computed yet7Sibling on the lefta5c3ca60d840719b…6d522d335338Parent hashNot computed yet8Sibling on the left39c4f42f564b0cdf…4fb5144380c8Parent hashNot computed yet9Sibling on the left4ed00875fe98b102…007bb7498c8bParent hashNot computed yet10Sibling on the rightd836050dd1857e2d…4ac62ccf3c8fParent hashNot computed yet11Sibling on the leftb67825d5230c9e40…9ef0b912e7a4Parent hashNot computed yet12Sibling on the rightf8106f1bd0687216…2964cb294092Parent hashNot computed yetComputed rootClimb all 12 levels to compute itMerkle root in the headerfaeda15aa61089ee792d9f53fc356561e7a8cc0936b371f8fd8248088ed25985Start at the transaction ID and climb one level at a time.
Proof size384 bytes12 sibling hashes of 32 bytes
Header plus proof464 bytesThe 80-byte header and the branch
Whole block1,681,649 bytes3,624 times the header and proof

The 12 sibling hashes total 384 bytes and link this transaction to its block, showing sides and parent hashes.

The header and proof together occupy 464 bytes, compared with 1.68 MB for the entire block.

The proof establishes inclusion only. It cannot establish transaction validity or reveal other transactions a server omitted.

After 12 double SHA-256 steps, the computed root matches the Merkle root stored in the header.

Changing one transaction ID byte changes every hash above it, so the computed root no longer matches.

Block, transaction and branch data come from mempool.space, 2 November 2026, checked by the page team through hashing.

BIP 37 introduced the Bloom filter approach in October 2012, letting wallets request matching transactions. A September 2014 study found that clients risked exposing almost all their addresses. BIP 157 later described most implementations as offering virtually zero privacy. Bitcoin Core 0.19.0.1 disabled serving Bloom filters by default in November 2019 to remove known denial-of-service vectors.

BIP 157 and BIP 158 define one compact block filter per block, checked locally. One honest peer lets clients identify correct filters, then fetch matching blocks from any peer. Filters are not committed in blocks. The Lightning node LND can use the Neutrino light client as its back end. Wasabi Wallet 2.8.0 obtains filters over P2P in June 2026. Electrum instead maintains about 10 server connections for headers. One main server sees wallet addresses and can lie by omission.

How Ethereum light clients work

LES stopped working after the Merge upgrade, and Geth removed it in November 2023. Ethereum's Beacon Chain instead supports the sync committee introduced with Altair on 27 October 2021. Each period assigns 512 seats for 8,192 slots, about 27.3 hours. Advancing a finalized header normally requires at least 342 signatures. A light client can follow period updates rather than downloading every block summary.

Figure 2 · Beacon API readings, 2 November 2026See sync committee participationSwitch 2 November 2026 samples from period 1,900, then adjust the signature count to see which accepted view can advance.
Sample
510 of 512
The 512 seats of the sync committeeSignedDid not signNot signing, what-ifSignatures for the attested header at slot 15,568,436, carried in the block at slot 15,568,437Signatures carried in the block at slot 15,568,436, 18:47:35 UTCSignatures carried in the block at slot 15,568,435, 18:47:23 UTCSignatures carried in the block at slot 15,564,801, the second slot of the period, 06:40:35 UTC
What the light client may acceptFinalized header may move510 of 512510 signatures reach the 2/3 rule of 342, so the finalized header may move if the other checks pass.Period 1,900: 06:40 UTC on 2 November to 09:58 UTC on 3 November 2026
Signed99.6%510 of 512 seats
Needed to finalize342Two thirds of 512, rounded up
Needed for the optimistic header256More than half of the best 510

The sampled finality update has 510 of 512 signatures for the attested header at slot 15,568,436.

With at least 342 valid signatures, the finalized header may advance when the other update checks pass.

With 256 to 341 signatures, participation exceeds half the recent best of 510, so only the optimistic header may advance.

At 255 signatures or fewer, neither view advances. The client waits for sufficient participation.

Signature bits come from public beacon API nodes, 2 November 2026. Thresholds follow the Altair light client specification.

Starting from a trusted root, the client verifies a BLS aggregate signature over the attested header. Merkle branches connect it to finality, the next committee and the execution payload header. Its state root anchors blockchain state, with eth_getProof returning Merkle Patricia Trie account and storage proofs. EIP-6800 estimated in March 2023 that account proofs average close to 3 kB. A finality update is about 2.1 kB. A full update each period is about 26.9 kB, roughly 8.6 MB yearly. The light client checks these connections without executing transactions.

Ethereum clients on 2 November 2026 include Helios 0.12.0, Nimbus's BETA light client, Lodestar 1.43.0, Geth blsync and Colibri 3.0.3. Lodestar's light client moved to its own repository in May 2026. ethereum.org lists all but Colibri and calls none production-ready. Portal Network remains unfinished. Weak statelessness remains research work. Most Verkle EIPs are Stagnant, with no stateless client EIP scheduled for a fork. Glamsterdam replaces the execution payload header with its block hash on Sepolia and Hoodi, requiring a separate execution header fetch. No mainnet date is set.

What clients download, check and trust

Storage volume does not identify what a program validates. This comparison separates downloaded material, local checks and outside assumptions. A pruned full node still validates blocks, while a light client follows selected consensus evidence. Requirements and implementation details below reflect 2 November 2026.

ClientDownloads or storesChecks itselfTrusts
Bitcoin full nodeDownloads or storesFull blocks, with at least 550 MiB retained when prunedChecks itselfValidates blocks and transactions under Bitcoin consensus rulesTrustsIts validation rules, not a server's transaction claims
Bitcoin SPV clientDownloads or storesEvery 80-byte header plus requested transactions and Merkle branchesChecks itselfValidates chain work and transaction inclusion against accepted rootsTrustsHonest hash power and servers for omitted activity
Neutrino clientDownloads or storesHeaders, compact filters and blocks matching wallet activityChecks itselfChecks filter consistency across peers and matching block dataTrustsHonest hash power and at least one honest filter peer
Ethereum full nodeDownloads or storesBlocks and state, potentially retaining only 128 recent state versionsChecks itselfExecutes transactions and enforces Ethereum consensus rules independentlyTrustsIts validation rules and any trusted recent starting block
Ethereum light clientDownloads or storesCommittee updates, requested account data and authentication branchesChecks itselfVerifies committee signatures, finality and requested values against rootsTrustsA recent checkpoint and an honest sync committee supermajority
Celestia light nodeDownloads or storesBlock summaries and 16 random samples per blockChecks itselfChecks data availability sampling results within a seven-day windowTrustsConsensus and sampling assumptions, without executing every transaction
ZK light clientDownloads or storesSuccinct consensus proofs for verification on a destination chainChecks itselfVerifies source consensus through a cryptographic verifier contractTrustsSource consensus majority, verifier correctness and, for SP1 Helios, a guardian

What a light client cannot catch

A valid light client response does not establish that a server supplied everything. Compare four readers here, from a single-server Bitcoin wallet to Ethereum committee verification. The results distinguish false data, missing data and delayed data, rather than treating every server lie alike.

Figure 3 · Sourced outcomes for four kinds of clientTest a server's claimChoose a false claim and compare which clients catch it, partly detect it or accept it without independent verification.
The server says
Bitcoin SPV, one serverCaughtThe Merkle branch does not reach the header's root
SPV with several peers or BIP 157CaughtThe Merkle branch does not reach the root
Ethereum sync committee clientCaughtThe receipt or account proof fails against the header
Wallet that trusts its RPC endpointNot caughtThe wallet takes the answer as it comes
Bitcoin SPV, one serverNot caughtA server can lie by omission
SPV with several peers or BIP 157CaughtOne honest peer serves the right filters
Ethereum sync committee clientNot caughtLog queries lack an efficient completeness proof
Wallet that trusts its RPC endpointNot caughtNothing is checked
Bitcoin SPV, one serverNot applicableBitcoin wallets track their own outputs
SPV with several peers or BIP 157Not applicableBitcoin wallets track their own outputs
Ethereum sync committee clientCaughteth_getProof fails against the state root
Wallet that trusts its RPC endpointNot caughtNothing is checked
Bitcoin SPV, one serverPartly caughtFake headers fail, but one server can hide more work
SPV with several peers or BIP 157CaughtProof of work and the most-work chain
Ethereum sync committee clientCaught342 of 512 committee signatures are needed
Wallet that trusts its RPC endpointNot caughtNothing is checked
Bitcoin SPV, one serverPartly caughtOnly header times give a hint
SPV with several peers or BIP 157CaughtOther peers report a newer chain
Ethereum sync committee clientPartly caughtThe client compares slots with its own clock
Wallet that trusts its RPC endpointNot caughtOld data comes back with no error
Bitcoin SPV, one serverNot caughtSPV trusts the hash-power majority
SPV with several peers or BIP 157Not caughtFilters do not check validity
Ethereum sync committee clientNot caughtThe client trusts 2/3 of the committee
Wallet that trusts its RPC endpointNot caughtNothing is checked

A fabricated transaction fails the Merkle proof for verifying clients. An RPC-only wallet accepts the server's claim.

One server can hide a transaction. BIP 157 exposes omissions with one honest peer, while Ethereum log queries lack efficient completeness proofs.

An incorrect Ethereum balance fails the account proof against the accepted state root. Bitcoin wallets instead track their outputs.

Fake headers fail proof of work or committee checks. A one-server SPV wallet cannot see a heavier chain, and an RPC-only wallet checks nothing.

Clocks only partly expose an old head. Additional peers can reveal fresher data that a single server withholds.

No light client here catches an invalid block accepted by a dishonest consensus majority, because none executes its transactions.

Outcomes follow the Bitcoin white paper and developer guide, BIP 157, Ethereum light client specifications, EIP-1186 and EIP-7745.

Weak subjectivity means a proof-of-stake client needs a recent trusted checkpoint to resist a long-range attack. Electra's window is about 15.7 days, while Gloas rules shorten it to about seven days. Helios warns after two weeks, and Nimbus requests a root about one to two weeks old. CometBFT explains the danger: old, unbonded voting keys can be bought to fabricate historical forks without current bonding costs.

Light clients across chains and bridges

Not every light client follows each intervening block. Some verify changing validator sets, some sample whether data is available, and others verify compressed computations. These differences matter when another blockchain, rather than a wallet, performs the checks.

CometBFT checks signatures exceeding 2/3 of the new set's voting power. Bisection skips intermediate headers but requires older-set support, defaulting to 1/3 in CometBFT and 2/3 in Hermes. The trusting period must also remain shorter than the unbonding period. In IBC, proven misbehavior freezes the client. Elsewhere, Celestia checks availability with 16 random samples per block within seven days. NEAR follows epoch block producers with approval above 2/3 of stake. Mina instead uses recursive proofs, with about 22 KB for non-consensus nodes.

A cross-chain bridge can verify source consensus rather than accept an oracle's claim. SP1 Helios is a ZK light client that checks source consensus inside the destination chain. Its Groth16 proofs are about 260 bytes, costing roughly 270,000 gas to verify. PLONK uses about 868 bytes and 300,000 gas. A guardian can change its verification key unless that role is relinquished. For blockchain rollups, ethereum.org describes checking deposit evidence before releasing tokens.

Light client risks to account for

Small downloads do not remove dependencies on peers, signers or verification code. These six failure modes affect different layers of a design. Separating them helps teams distinguish an unavailable answer from an answer that should never be accepted.

Lies by omission

A server can hide activity without fabricating anything. Ethereum has no efficient log index in November 2026, so proving that a log query is complete remains expensive.

Eclipse and stale data

An eclipse attack isolates a reader. Multiple peer-to-peer connections help, but partitioning or Sybil identities can defeat them, leaving valid yet outdated responses.

Old checkpoints

A stale checkpoint can anchor a convincing historical fork. Proof-of-stake readers need a recent trusted starting point, not merely internally consistent signatures.

Unslashable committee signatures

Sync committee members cannot be slashed for signing a wrong root. EIP-7657 warns that a dishonest supermajority can present non-canonical finality.

Server privacy leaks

Electrum servers learn script hashes and IP addresses. BIP 157 sends no wallet data, but Ethereum RPC providers still receive every query.

Bridge verifier bugs

Wormhole in February 2022, Nomad in August 2022 and BNB Bridge in October 2022 lost funds through broken verification. Their consensus mechanisms did not fail.

How light clients developed

The design began with payment inclusion, then expanded to address privacy and different consensus systems. This timeline follows eight protocol and implementation changes. It separates proposed ways to obtain data from the software releases that made them available.

  1. October 2008

    Payment verification without full blocks

    The Bitcoin white paper introduces Simplified Payment Verification in section 8. Readers can connect payments to the chain without executing transactions.

  2. October 2012

    Bloom filters enter SPV

    BIP 37 lets peers select transactions matching a client's Bloom filter. The client no longer needs every transaction to find possible matches.

  3. May 2017

    Compact filters change requests

    BIP 157 and BIP 158 define filters that clients download and match locally, rather than sending wallet-specific filters to peers.

  4. November 2019

    Bloom serving becomes optional

    Bitcoin Core 0.19.0.1 stops serving Bloom filters by default. The change removes well-known denial-of-service vectors from the default configuration.

  5. October 2021

    Altair introduces committee signing

    Ethereum adds sync committees on 27 October 2021. Their signatures let constrained readers follow consensus without processing every transaction themselves.

  6. November 2023

    Geth removes broken LES

    Geth removes its LES code on 23 November 2023 after it stops working following the Merge. Version 1.13.6 follows in December 2023.

  7. June 2026

    Wasabi moves filter retrieval

    Wasabi Wallet 2.8.0 obtains compact filters from the Bitcoin P2P network. This replaces its centralized server for that data.

  8. October 2026

    Clients and formats change

    Helios 0.12.0 is released. Glamsterdam changes light client headers on Sepolia on 6 October 2026 and Hoodi on 26 October 2026.

Planning light client checks

Adding verification to a wallet, app or bridge starts with boundaries, not a client name. We separate what the application can verify from what it still assumes. In crypto wallet development, that means deciding how data reaches the user, how it is authenticated and what happens when verification cannot finish.

PixelPlex case icon

PixelPlex was founded in 2007 and began blockchain work in 2013. Our Qtum Wallet case study covers Android and iOS, Qtum API integration, on-device encoding of private keys and all transaction data, and a 12-word seed phrase backup. This demonstrates wallet integration experience, not a light client implementation.

  1. Root of trust

    We identify the consensus anchor and checkpoint source first. The application needs a defined response when that starting point becomes too old.
  2. Data to verify

    We distinguish balances, transaction inclusion and logs. Each requested value needs an applicable verification path, while missing activity needs separate treatment.
  3. Peers and providers

    We map which peers or RPC providers supply each answer. Extra sources can expose withheld updates, but cannot replace consensus validation.
  4. Query privacy

    We record what requests reveal, including addresses and network activity. Verification protects against some false answers, not against a provider observing queries.
  5. Fork compatibility

    We track format changes such as Glamsterdam's separate execution header fetch on test networks in October 2026, keeping verification boundaries explicit.

Discuss verification boundaries for your next application

We can discuss your data sources, verification requirements and remaining assumptions before you choose an implementation approach.

Discuss your project

Light client FAQ

What is a light client in simple terms?

A light client is software that follows a blockchain using headers or committee updates and checks the data it requests. It relies on others for full transaction validation. The term light node refers to a node running this software, not a separate consensus role.

What separates a light client from a full node?

A full node validates transactions and enforces the chain's rules, while a light client checks selected consensus evidence and requested data. Pruning does not change that distinction. A Bitcoin full node can retain as little as 550 MiB of blocks.

What does SPV mean in Bitcoin?

SPV means Simplified Payment Verification, the method described in section 8 of the October 2008 white paper. It follows the most-work header chain and connects a confirmed transaction to a block using a Merkle proof. It does not independently validate that transaction.

How does an Ethereum light client authenticate headers?

An Ethereum light client starts from a trusted checkpoint and verifies the sync committee's aggregate BLS signature, along with the update's required Merkle branches. Finalized progress needs 342 of 512 signatures. It also checks that updates do not skip a committee period.

Is a light client as secure as a full node?

No. A light client checks less and depends on honest consensus participation, while a full node independently rejects blocks that break its rules. Security also depends on freshness and verifier correctness. A valid inclusion result does not establish that the server disclosed every relevant transaction.

Can light clients run on phones or in browsers?

Light clients can run on phones, an intended environment for constrained verification software. On 2 November 2026, Helios compiles to WebAssembly, and Nimbus reports under 1 MB of disk while labeling its client BETA. These resource properties do not establish production readiness.

How do bridges use light clients?

A bridge can verify deposits when the destination chain runs a light client of the source chain. This moves the decision away from simply trusting an oracle. The verifier still has to implement every required check correctly. Administrative powers, such as changing verification keys, remain separate assumptions.
Sources (35)

Get updates about blockchain, technologies and our company

We will process the personal data you provide in accordance with our Privacy policy. You can unsubscribe or change your preferences at any time by clicking the link in any email.

Follow us on social networks and don't miss the latest tech news

  • facebook
  • X
  • linkedin
  • instagram
Stay tuned and add value to your feed