pracadata/brazil-fuel-prices

Brazil Fuel Prices by Municipality (weekly, ANP)

Retail fuel prices in Brazil, by municipality, product and week, from the weekly price survey of the national oil agency (ANP): gasoline, premium gasoline, ethanol, diesel (and diesel S10), CNG (GNV) and LPG cooking gas, 2004-05-10 to 2026-10-12 (12 October 2026). One row per municipality × product × week with the number of stations and, when there are at least 3 stations, the mean, median, minimum, maximum and quartiles of the pump price in BRL. A second table says, for every week of the calendar, how complete the survey was. Updated weekly. The latest week in this build is 2026-10-12.

Per-station prices and ethanol/gasoline parity: on the Apify Store. Pay per row, in BRL, USD and EUR, with an API and MCP for AI agents. The companion dataset pracadata/brazil-vehicle-fleet has the registered vehicle fleet by municipality and fuel type (free).

Read this first: coverage and breaks

• The ANP survey is a sample, not a census. It covers about 400 municipalities in the latest week (Brazil has 5,571) and 893 municipalities over the whole history. Prices are for the stations the ANP surveys, not for every pump.
• The panel shrank in July 2024 (from about 430 to about 345 municipalities, and from about 9,300 to about 5,600 stations in a week) and again between April and July 2025 (minimum 285 municipalities in the week of 14 July 2025). Series by municipality can disappear or reappear because of it; national averages mix a different set of municipalities before and after. Both are marked in break_event_id and are not hidden.
• Four documented breaks (municipalities in the survey per week, gasoline or ethanol unless noted):

break_event_id	Weeks	What happened	Effect on coverage_flag
collection_pause_2020	2020-08-17 to 2021-05-17	ANP paused collection from 23 Aug 2020. Diesel, CNG and LPG resumed on 19 Oct 2020 with 29 municipalities; gasoline and ethanol resumed on 28 Dec 2020 and stayed under 250 municipalities until 24 May 2021 (461 before the pause)	partial or gap
outage_2022_09	2022-09-12 to 2022-12-19	Collection almost stopped: 413 municipalities (12 Sep), 1 (19 Sep), 27 (26 Sep, 3 Oct), 104 (10 Oct), then a slow recovery; last week under 250 was 19 Dec 2022	partial
coverage_cut_2024_07	from 2024-07-01	Coverage dropped by about a fifth: 428 municipalities (24 Jun 2024) to 343 (1 Jul 2024); stations 9,270 to 5,643	note only
coverage_cut_2025	2025-04-07 to 2025-07-28	Further decline: 382 (28 Apr 2025), 368 (7 Jul), 285 (14 Jul), 275 (28 Jul); it recovers to about 380–400 in 2026	note only

• coverage_flag describes the week, not the series: complete (250 municipalities or more with gasoline or ethanol, outside a pause or outage), partial (data, but below 250 municipalities or inside a pause/outage) and gap (a calendar week with no data at all). In this build: 687 complete, 405 partial and 79 gap weeks. in_continuous_series is true from 2022-12-26 on: from that week the survey has no missing weeks. Before it, even complete weeks are surrounded by gaps (the ANP files for several semesters are partial), so filter on in_continuous_series if you need an uninterrupted weekly series.
• Sale price only. The ANP also reported a purchase (distributor) price until 17 August 2020; it is not included.
• Brazilian reais, nominal. No USD/EUR conversion and no inflation adjustment in this dataset (the per-station Actor on Apify has USD and EUR).
• At least 3 stations to publish a statistic. With fewer, only n_stations is filled and the price columns are null.

What is in this repository

File (config)	Rows	Content
data/fuel_price_municipality_weekly.parquet (fuel_price_municipality_weekly)	1,887,188	municipality (code, name and state) × product × week, 16 columns
data/fuel_coverage_weekly.parquet (fuel_coverage_weekly)	1,171	one row per calendar week (Mondays) with the survey size and coverage_flag, 7 columns
DATA_DICTIONARY.md		every column: type, definition, where it is null; products and units
manifesto.json		files, row counts, SHA-256 of each file, date of the data build, sources

Products: gasoline, gasoline_premium (from December 2020), ethanol, diesel, diesel_s10 (from December 2012), diesel_s50 (2012 only), cng (price per m³) and lpg (price per 13 kg cylinder). All others are BRL per litre. unit says which.

Examples

Ethanol and gasoline in the latest week, average of municipality medians by state (the state column has the two-letter state code; the first two digits of ibge_code are the IBGE state code: 35 São Paulo, 33 Rio de Janeiro, 31 Minas Gerais), in DuckDB:

with p as (
  select * from 'hf://datasets/pracadata/brazil-fuel-prices/data/fuel_price_municipality_weekly.parquet'
  where week_start = date '2026-10-12' and product in ('ethanol', 'gasoline') and price_brl_median is not null
)
select substr(ibge_code, 1, 2) as state_code,
       round(avg(price_brl_median) filter (where product = 'ethanol'), 2) as ethanol_brl_l,
       round(avg(price_brl_median) filter (where product = 'gasoline'), 2) as gasoline_brl_l,
       round(avg(price_brl_median) filter (where product = 'ethanol') / avg(price_brl_median) filter (where product = 'gasoline'), 3) as ratio
from p group by 1 order by ratio;

A ratio at or below 0.70 is the usual rule of thumb for ethanol being cheaper per kilometre in a flex-fuel car. The same in pandas, national median by week since the series became continuous (needs pyarrow and huggingface_hub installed):

import pandas as pd
df = pd.read_parquet("hf://datasets/pracadata/brazil-fuel-prices/data/fuel_price_municipality_weekly.parquet")
g = df[(df["product"] == "gasoline") & df["in_continuous_series"]].groupby("week_start")["price_brl_median"].median()
print(g.tail())

from datasets import load_dataset
ds = load_dataset("pracadata/brazil-fuel-prices", "fuel_coverage_weekly")

Source and attribution

• Fonte: ANP, Série Histórica de Preços de Combustíveis. Levantamento de Preços de Combustíveis of the Agência Nacional do Petróleo, Gás Natural e Biocombustíveis, open data files: https://www.gov.br/anp/pt-br/centrais-de-conteudo/dados-abertos/serie-historica-de-precos-de-combustiveis.
• Fonte: IBGE, Localidades. Municipality codes: https://servicodados.ibge.gov.br/api/v1/localidades/municipios.
• Exchange rates (used only in the per-station Actor on Apify): European Central Bank euro reference rates.
• Licence and terms of use. The licence field is other because the data is public data of the Brazilian government, not under a dataset-specific licence. The Terms of Use of the ANP portal (https://www.gov.br/anp/pt-br/acesso-a-informacao/termos-de-uso-privacidade-e-seguranca-1) authorise "a reprodução total ou parcial, sem fins lucrativos, do conteúdo, mediante a citação obrigatória da fonte". This dataset is free and cites the source above. The open-data policy of the Brazilian federal executive (Decreto 8.777/2016) also defines open data as free to use with credit to the source. This dataset is a derived, unofficial work and is not affiliated with, endorsed by or approved by the ANP. Check the source terms for your use.
• Takedown / corrections: write to lucashpissaia@gmail.com. The data has no personal information in this repository (prices aggregated by municipality; station names and addresses are not published here).

How to cite

@dataset{brazil_fuel_prices,
  title  = {Brazil Fuel Prices by Municipality (weekly, ANP)},
  author = {pracadata},
  year   = {2026},
  url    = {https://huggingface.co/datasets/pracadata/brazil-fuel-prices},
  note   = {Derived from ANP and IBGE open data. Fonte: ANP, Série Histórica de Preços de Combustíveis; Fonte: IBGE, Localidades}
}

Terms

• Provided as is, without warranty, from public sources, with the limitations above. Not official; do not use as a regulated price reference without checking the ANP.
• Cite the sources above. If a publisher restricts reuse, the data may be removed from this repository going forward.
• Questions or corrections: open a discussion on this repository. Takedown requests: lucashpissaia@gmail.com.