The best exchange rate datasets in 2026: 8 compared

Disclosure: we publish one of the datasets below. We checked the others against their own pages on 6 November 2026, and where a page did not state something, we say so rather than fill the gap from memory. Terms change, so read the current licence before you ship.

The eight datasets at a glance

What makes a good exchange rate dataset

Coverage is the number everyone quotes, but four other questions decide whether a dataset works for you.

• Who produced the number? A rate printed by a central bank can be cited. A blended or computed rate might be better for analysis, but an auditor cannot trace it to one publisher.
• Is the value a real rate? Some machine learning datasets rescale their values. They train models well and convert nothing.
• What does the date mean? A period average, an end-of-period rate, a noon buying rate and a daily fixing are four different things, even for the same pair and month.
• What may you do with it? A dataset that is free to download is not always free to use in a product.

1. AllRatesToday central bank exchange rates

Best for official rates by institution. Our dataset holds the rates that 123 central banks and 6 tax authorities published, more than 15 million rows in all, with history back to 1914 where records exist. Every row is a figure the institution printed, in the direction it printed it. We do not compute cross rates in the dataset, so you can trace a row to its publisher.

It is also the most downloaded dataset on Hugging Face for the search "exchange rate": 9,784 downloads in the last 30 days when we checked on 6 November 2026, against 595 for the next result, and 15,216 all time. A download there is a file request, not a person, so read what the counter means before quoting it.

There are real limits. The licence is CC BY-NC 4.0, so commercial use needs the Official Rates plan. Publishers differ in what they print and how they label it, so a "buy" rate from one bank may not match another's. The Hugging Face copy trails by two days and needs a free login and a free API key. The GitHub repository needs no account and carries the latest tables.

import pandas as pd

# No account needed: one year of ECB rates from GitHub
url = ("https://raw.githubusercontent.com/AllRates-Today/"
       "central-bank-exchange-rates/main/data/ecb/history/2026.csv")
df = pd.read_csv(url)  # date, base, quote, type, value

2. BIS bilateral exchange rates

Best for one consistent cross-country series. BIS offers daily series for around 80 economies and monthly, quarterly and annual series for about 190. Most daily series start around 1970, 14 currencies go back to 1950, and lower-frequency series start around 1957. It has bulk downloads and an SDMX API, and it updates weekly.

Use of the statistics is unrestricted on conditions: cite BIS, do not mislead, and do not charge subscribers extra because a commercial publication includes them. BIS is the first place to look for macro research. It is a poor fit if you need the current week's rate.

3. IMF Exchange Rates

Best for SDR rates and period averages. The IMF dataset covers the US dollar, SDR, euro and national currencies, with both period-average and end-of-period rates, which saves you computing them. You can download it or use the API and an Excel add-in.

Read the terms closely. The data terms allow reuse with attribution, but they send commercial reuse to copyright@imf.org for permission. The general terms also prohibit automated bulk download and LLM training without permission. Fine for a paper; check first for a pipeline.

4. World Bank official exchange rate

Best for commercial use of yearly figures. Indicator PA.NUS.FCRF gives the official exchange rate in local currency per US dollar as a period average, in CSV, XML or Excel. Its CC BY 4.0 licence allows commercial use with attribution, the clearest terms in this list.

The World Bank credits the IMF's International Financial Statistics as the source, so this is a repackaging of IMF data rather than an independent series. Its page does not say whether IMF terms also apply, so check if that matters to you. It suits country-level economics, not invoice-date conversion.

5. ECB euro reference rates

Best for clean daily euro rates. The ECB publishes 29 currencies against the euro each working day, usually around 16:00 CET, as CSV, XML and through its data portal. The files are small, regular and easy to automate. The EUR/RUB rate has been suspended since March 2022.

The ECB says the rates are for information only and strongly discourages using them for transactions. Everything is quoted against the euro, so any other pair is your own calculation.

6. Federal Reserve H.10

Best for US dollar noon buying rates. H.10 gives daily rates and dollar indexes, released on Mondays at 4:15 p.m. for the previous business week. The rates are noon buying rates in New York for cable transfers, certified by the New York Fed for customs and SEC purposes.

If you automate this one, note a change. The Fed says it will remove the "Build Your Package" feature of its Data Download Program during the week of 9 November 2026, ahead of retiring the program, and points users to FRED. It also notes that weekly publication stopped between May 2006 and January 2009, with data for that gap held in the download program.

7. Frankfurter

Best for a free blended rate. Frankfurter is an open-source API, not a file archive, but it is a practical way to pull history. It states 219 currencies from 108 central banks and official sources, history back to 1948, and JSON, CSV or NDJSON output.

Its blended rates are its own calculation and free for commercial use, a real advantage over our licence. The trade-off is the one in the first question above: a blended figure is not a number a publisher printed. Its direct provider endpoints stay under each provider's terms. We compare the two in detail in AllRatesToday vs Frankfurter.

8. HistData

Best for free intraday bars. HistData is a different kind of dataset: tick and one-minute data for forex pairs, metals, indices and commodities, as zipped CSVs by pair, year and month, in MetaTrader, NinjaTrader, Excel and generic formats. It is for backtesting a trading strategy, not for accounting.

Bars are built from bid prices, and timestamps are Eastern Standard Time with no daylight saving adjustment, which will catch you out if you join them to other data. We found no licence on the pages we read, only a no-warranty notice, so ask before using it in a product.

One to avoid for real conversions

Search Hugging Face for "exchange rate" and you will find ts-arena/exchange_rate and copies of it. It is a good forecasting benchmark, with daily data for eight countries from 1990 to 2016, but its values are normalised to zero mean and unit variance. Use it to compare models. Do not use it to convert anything.

Which one should you pick?

Frequently asked questions

What is the best free exchange rate dataset?

It depends on the job. For official rates from individual central banks and tax authorities, use the AllRatesToday central bank dataset. For one consistent series across about 190 economies, use the BIS bilateral exchange rates. For yearly figures under a licence that allows commercial use, use the World Bank indicator. For intraday price bars, use HistData.

Which exchange rate dataset has the longest history?

Of the eight here, the AllRatesToday dataset reaches furthest, with rows back to 1914 for the institutions that have records that old. BIS has 14 currencies back to 1950, and Frankfurter states history back to 1948. Coverage that old is thin in every dataset, so check the specific currency before relying on it.

Can I use these exchange rate datasets commercially?

The terms differ. The World Bank indicator is CC BY 4.0, which allows commercial use with attribution. BIS allows use with citation, and commercial publications may not charge extra for including its statistics. The IMF asks you to request permission for commercial reuse. The AllRatesToday dataset is CC BY-NC 4.0, so commercial use needs the Official Rates plan. Read the current terms before you ship anything.

Is the Hugging Face exchange_rate dataset real exchange rate data?

The ts-arena copy is a machine learning benchmark: eight countries from 1990 to 2016, standardised to zero mean and unit variance. Its values are not rates you can quote or convert with.

Do these datasets include today's rate?

Most trail in some way. BIS updates weekly, the Federal Reserve publishes the previous week on Mondays, and our Hugging Face copy trails by two days. The ECB publishes each working day. For a current table, use an API or the source's daily file.