Tools/Explorers

Bitcoin Fee Estimation API Accuracy Comparison

Compare Bitcoin fee estimation APIs by prediction accuracy, update frequency, mempool data sources, and API reliability for 1-block, 3-block, and 6-block targets.

Spark Team

How Accurate Are Bitcoin Fee Estimation APIs?

Choosing the wrong fee estimation API can cost you money in two ways: overpaying wastes sats, and underpaying leaves your transaction stuck in the mempool for hours or days. The gap between providers is larger than most developers expect. In Block's Augur benchmark (May 2025), next-block miss rates ranged from 3.6% to 24.4%, while average overpayment ranged from 15.9% to 65.5% across the same providers.

This comparison covers the most widely used public Bitcoin fee estimation APIs: mempool.space, Blockstream Esplora, BitGo, Bitcoin Core's estimatesmartfee RPC, Bitcoiner.live, and the legacy Earn.com endpoint. Each uses a different estimation methodology, and those differences determine how they perform during fee spikes, congestion, and quiet periods.

Provider Overview

The following table summarizes the key characteristics of each fee estimation API. Methodology is the most important differentiator: mempool-based estimators simulate the next block template from pending transactions, while historical estimators analyze confirmation data from past blocks.

ProviderMethodologyEndpointAuth RequiredUpdate FrequencySegWit/Taproot Aware
mempool.spaceMempool-based/api/v1/fees/recommendedNo~2 secondsYes (vBytes)
Blockstream EsploraHistorical (Core-based)/fee-estimatesNo~60 secondsYes (sat/vB)
BitGoProprietary hybrid/api/v2/btc/tx/feeYes (API key)Per-requestYes (sat/kB)
Bitcoin CoreHistorical bucket-basedestimatesmartfee RPCLocal nodePer-blockYes (BTC/kvB)
Bitcoiner.liveMempool-based/api/fees/estimates/latestNo~30 secondsYes (sat/vB)
Earn.com (legacy)Historical/api/v1/fees/recommendedNoDeprecatedNo (sat/byte)

Earn.com's fee API (formerly bitcoinfees.earn.com) is effectively deprecated. Some wallet libraries still reference it, but the endpoint has not been reliably maintained. Blockchain.com's fee API is similarly undocumented for public use. Neither should be relied on for production applications.

Accuracy Benchmark: Next-Block Estimation

The most rigorous public benchmark comes from Block's Augur project, which tested fee estimation providers during May 2-15, 2025 against actual confirmation data for a 1-3 block target. The benchmark compared each provider's estimate at the time of broadcast against the fee rate that actually confirmed in subsequent blocks.

ProviderMiss RateAvg OverpaymentNet Cost DifferenceInterpretation
Bitcoiner.live3.6%65.5%63.2%Rarely misses, but severely overpays
Augur (Block)14.1%15.9%4.5%Best balance of reliability and cost
WhatTheFee14.0%18.7%7.7%Competitive accuracy and cost
Blockstream Esplora18.7%44.2%31.9%Moderate miss rate, high overpayment
mempool.space24.4%21.7%3.0%Higher miss rate, but low cost when it hits

These results reveal a fundamental tradeoff. Bitcoiner.live achieves near-zero misses by quoting aggressively high fees: users pay roughly 65% more than necessary. Mempool.space takes the opposite approach: its estimates are closer to the minimum viable fee, which keeps costs low but means more transactions miss their target block. Blockstream Esplora, which wraps Bitcoin Core's historical estimator, sits in an uncomfortable middle: a meaningful miss rate combined with significant overpayment.

For a deeper analysis of how these algorithms work internally, see our Bitcoin fee estimation algorithm comparison.

Mempool-Based vs. Historical Estimation

The core architectural difference between fee estimators is their data source. This distinction explains most of the accuracy differences observed in benchmarks.

Mempool-Based Estimators

Mempool.space and Bitcoiner.live simulate what the next block template would look like by sorting all pending transactions by fee rate and filling a virtual block. Mempool.space runs this simulation approximately every two seconds and returns four tiers: fastestFee, halfHourFee, hourFee, and economyFee. It also exposes a /api/v1/fees/mempool-blocks endpoint that returns projected block contents with fee ranges, giving developers fine-grained data to build their own estimation logic.

The strength of mempool-based estimation is responsiveness. When a wave of high-fee transactions enters the mempool during an Ordinals mint or a market panic, the estimator reflects the new reality within seconds. The weakness is that it only sees transactions in the local node's mempool, which may diverge from what miners actually select. Policy differences, private transaction relay, and replace-by-fee activity all create discrepancies.

Historical Estimators

Bitcoin Core's estimatesmartfee RPC uses a confirmation-history approach. It groups past transactions into fee-rate buckets and tracks what percentage of each bucket confirmed within N blocks, requiring 95% confidence before returning an estimate. It offers two modes: conservative (uses the higher of short- and long-window estimates) and economical (uses the lower).

Blockstream's Esplora API wraps this same methodology, returning a map of confirmation targets (2, 3, 4, 6, 10, 20, 144, 504, 1008 blocks) to fee rates in sat/vB. Because both rely on historical data, they share a critical limitation: during rapid fee changes, the estimate reflects where fees were, not where they are. Analysis from Bitcoin Core contributor ismaelsadeeq demonstrated this at block 812,595, where the historical estimator suggested 18 sat/vB while the mempool-based estimate was 33 sat/vB. Transactions at the historical rate remained unconfirmed after multiple blocks.

In Bitcoin Core's conservative mode, the problem compounds. Analysis of block 847,088 found the conservative estimate at approximately 506 sat/vB while the economical mode returned roughly 48 sat/vB. That 10x gap represents the cost of maximum safety margins in a historical model.

API Response Formats and Integration

Each provider returns fee data in a different format and unit. This inconsistency is a common source of bugs in wallet implementations.

ProviderUnitResponse FormatConfirmation Targets
mempool.spacesat/vB (integer)Named tiers: fastestFee, halfHourFee, hourFee, economyFee, minimumFee~1, ~3, ~6, ~144 blocks
Blockstream Esplorasat/vB (decimal)Map of target → fee rate2, 3, 4, 6, 10, 20, 144, 504, 1008 blocks
BitGosat/kBfeePerKb, feeByBlockTarget, numBlocks, confidence1-144 blocks with confidence %
Bitcoin CoreBTC/kvBfeerate, blocks (target), errors[]1-1008 blocks
Bitcoiner.livesat/vBEstimates by time window30 min, 60 min, 120 min, 360 min, 720 min
Unit conversion trap: BitGo returns sat/kB while most other APIs return sat/vB. To convert: sat/vB = sat/kB ÷ 1000. Bitcoin Core returns BTC/kvB, so multiply by 100,000,000 and divide by 1000 to get sat/vB. Getting this wrong will either massively overpay or create transactions that never confirm.

Behavior During Fee Spikes

Fee spikes are where estimation differences matter most. During fee spikes, the mempool fills rapidly and the minimum fee for next-block inclusion can jump from 5 sat/vB to 200+ sat/vB within minutes.

  • Mempool-based estimators (mempool.space, Bitcoiner.live) detect spikes within seconds because they rescan the pending transaction pool continuously. Mempool.space also offers a WebSocket endpoint at wss://mempool.space/api/v1/ws for real-time fee updates.
  • Historical estimators (Bitcoin Core, Blockstream Esplora) lag by multiple blocks because they need new confirmation data to update their statistical model. During the spike itself, they return stale estimates based on pre-spike conditions.
  • BitGo's proprietary estimator uses a hybrid approach. It includes mempool awareness but adds a safety margin that keeps estimates above the network minimum, which can result in overpayment during stable periods.

Developers building time-sensitive payment systems should consider blending multiple data sources. The bitcoin-blended-fee-estimator project by LN-Zap demonstrates this approach, querying mempool.space and Esplora in parallel and applying a weighting function.

SegWit and Taproot Fee Awareness

Modern fee estimators must account for the difference between raw bytes and virtual bytes. Since SegWit (2017), witness data is discounted at a 4:1 ratio, meaning a SegWit transaction's fee is calculated per virtual byte (vByte), not per raw byte. All current major APIs report in sat/vB or an equivalent unit that accounts for the witness discount.

Taproot (P2TR) transactions benefit from additional weight savings, particularly for key-path spends. When estimating fees for Taproot transactions, the fee rate from any of these APIs applies correctly because the wallet computes the transaction's virtual size using the Taproot weight rules, then multiplies by the sat/vB rate. The one exception is the legacy Earn.com API, which returned sat/byte (not sat/vB) and did not account for the witness discount. Using that rate for SegWit or Taproot transactions would result in overpayment.

For wallets that support both legacy and modern address types, see our fee estimator tool to compare the cost difference by transaction type.

Rate Limits and Reliability

For production deployments, API availability matters as much as accuracy. A fee estimator that goes down during a congestion event is worse than a slightly less accurate one that stays online.

  • mempool.space: approximately 10 requests per second on the public instance without an API key. Self-hosting removes all rate limits. The project is fully open source.
  • Blockstream Esplora: free public API with an estimated 500,000 requests per month. Also open source and self-hostable.
  • BitGo: requires an API key. Rate limits depend on the account tier. Enterprise customers get higher limits and SLA guarantees.
  • Bitcoin Core RPC: no external rate limits since it runs locally, but requires running a full node with an up-to-date mempool.
  • Bitcoiner.live: free public API, but uptime guarantees and rate limits are not formally documented.

Self-hosting mempool.space or Esplora eliminates the dependency on third-party uptime entirely. This is the recommended approach for exchanges, payment processors, and any application where transaction confirmation timing is business-critical.

Which API Is Best for Your Use Case

There is no single best fee estimator. The right choice depends on whether you prioritize low miss rates (reliability) or low overpayment (cost efficiency).

  • For time-sensitive payments (exchange withdrawals, merchant settlements): use mempool.space or a blended estimator. The low overpayment rate keeps costs down, and the real-time mempool data tracks fee changes within seconds.
  • For high-value transactions where confirmation certainty matters more than fee savings: Bitcoiner.live or BitGo's conservative estimates minimize the chance of a stuck transaction.
  • For applications with RBF support: start with mempool.space's economyFee and bump via RBF if the transaction does not confirm. This approach captures the cost savings of aggressive estimation while maintaining a fallback. See our fee bumping guide for implementation details.
  • For sovereign infrastructure: run Bitcoin Core with estimatesmartfee in economical mode. It avoids third-party dependencies entirely, though at the cost of slower reaction to fee changes.
  • For Layer 2 protocols: Lightning and Spark transactions settle off-chain and avoid on-chain fee estimation for most payments. When on-chain settlement is needed (channel opens, closes, or VTXO management), mempool-based estimation is preferred for its responsiveness.

Frequently Asked Questions

Why do fee estimates differ between providers?

Fee estimates differ because providers use fundamentally different methodologies. Mempool-based estimators (mempool.space) look at what is currently waiting to be mined and simulate block construction. Historical estimators (Bitcoin Core, Blockstream) look at what confirmed in recent blocks and extrapolate. Each approach has blind spots: mempool-based estimators cannot see private transaction relay or miner-specific ordering, while historical estimators cannot react to sudden demand changes until new blocks confirm.

Which API is best for time-sensitive payments?

For time-sensitive payments, mempool.space offers the best combination of real-time data and low overpayment. Its 2-second update cycle means estimates reflect current conditions almost immediately. For maximum reliability, combine it with RBF-enabled transactions so you can bump the fee if initial estimates prove too low. For payments on Bitcoin Layer 2 networks like Lightning or Spark, on-chain fee estimation is not needed for routine transfers.

Is Bitcoin Core estimatesmartfee accurate enough for wallets?

Bitcoin Core's estimatesmartfee is reliable during stable fee environments but struggles during volatility. Its conservative mode can overestimate by 10x or more during fee spikes because it relies on historical confirmation data. The economical mode is more practical for most wallet use cases. Many wallets supplement Core's estimates with mempool data from external APIs to improve accuracy during congestion.

How do I test fee estimation accuracy for my application?

Block's open-source augur-benchmarking tool provides a framework for backtesting fee estimators against historical block data. It records estimates from multiple providers and compares them against the fee rates that actually confirmed. The dataset covers January through June 2025. You can extend it to benchmark your own estimator or test provider accuracy over different time windows and network congestion levels.

Should I use sat/vB or sat/byte for fee estimation?

Always use sat/vB (satoshis per virtual byte). Since the SegWit upgrade, virtual bytes account for the witness discount that makes SegWit and Taproot transactions cheaper. Using raw sat/byte for a SegWit transaction means paying roughly 40-75% more than necessary. All modern fee APIs except the deprecated Earn.com endpoint report in sat/vB or an equivalent virtual-byte-based unit.

Can I combine multiple fee estimation APIs?

Yes, and many production systems do. A common pattern is to query two or three providers in parallel and take the median or weighted average. This reduces the impact of any single provider's blind spot. The bitcoin-blended-fee-estimator project implements this pattern. The tradeoff is added latency and complexity: you need fallback logic for when one provider is down, and you need to handle the unit conversion differences between APIs.

How often should my application poll a fee estimation API?

For most applications, polling every 30-60 seconds is sufficient. Fee rates rarely change meaningfully within shorter intervals except during extreme spikes. If you need real-time data, mempool.space offers a WebSocket connection that pushes updates as the mempool changes. For pre-signed or batched transactions, estimate the fee at signing time rather than relying on a stale cached value.

This comparison is for informational purposes only and does not constitute financial advice. Accuracy data is based on Block's Augur benchmark (May 2-15, 2025) and may not reflect current performance. Fee estimation accuracy varies with network conditions. Always verify current API documentation and test against live data before deploying in production.

Build with Spark

Integrate bitcoin, Lightning, and stablecoins into your app with a few lines of code.

Read the docs →