Open data: every Gambeta AI pick, with its result

• Two formats: JSON and CSV, with every pick and its result.
• CC BY 4.0 license: use it freely and credit Gambeta with a link.
• Recomputable: with the file's columns you get the same hit rate and yield we publish.

What this dataset is

It is the record of every pick published by the Gambeta AI.

Today it holds 1,720 picks, from 2026-03-21 to 2026-11-08, and 1,697 are already resolved.

Each pick is published before the match starts.

The result is filled in automatically, with data from independent public sources.

We publish it because the idea is simple.

We publish everything, the good and the bad, and anyone can check it.

A record you cannot download is a promise, not data.

Download: the two files

The full addresses start with https://gambeta.ai.

For example, the CSV is at https://gambeta.ai/data/gambeta-historial.csv.

Both files hold the same data.

Field dictionary

The column names are in Spanish, exactly as they appear in the file.

The news columns (noticias_u, noticias_origen and noticias_tier) show whether local press moved the pick.

An empty field means the data does not exist for that pick.

How to recalculate hit rate and yield

You can check our numbers in a few steps.

Hit rate is the percentage of picks won among those that were won or lost.

Yield is the total profit divided by the amount staked.

Here is an example in Python.

import pandas as pd
d = pd.read_csv("gambeta-historial.csv")
dec = d[d.resultado.isin(["win", "loss"])]
print("Hit rate %:", round(100 * (dec.resultado == "win").mean(), 1))
mes = d[d.fecha.str.startswith("2026-10") & d.resultado.isin(["win", "loss", "void"])]
print("October yield %:", round(100 * mes.ganancia.sum() / mes.stake.sum(), 2))

Each pick counts in the month of its match date.

With this code and the CSV, the October 2026 yield gives the same number we publish in the October report.

If the file changes because of a resolver correction, the number can move by a few tenths.

License and how to cite it

The dataset is published under the Creative Commons Attribution 4.0 license (CC BY 4.0).

You can copy it, redistribute it and use it in your own analyses, including commercially.

The only condition is to credit Gambeta and link to gambeta.ai.

If you modify it, you must say so.

You can use this citation:

Gambeta AI (2026). Verified AI picks track record [dataset]. gambeta.ai. https://gambeta.ai/data/gambeta-historial.csv. License CC BY 4.0.

What you can do with this data

• Evaluate an AI: compare the hit rate with the implied probability of the odds.
• Study variance: see how many losing streaks appear in a large sample.
• Teach statistics: use a real case with wins, losses and odds.
• Write a story: cite a data point with a source, without having to ask us.

Limits worth knowing

• It is not the whole market: only the picks our AI published are in it.
• Some picks are pending: matches that have not finished appear as pending.
• Corrections happen: if the resolver fixes a result, the file is updated.
• A monthly or per-sport sample can be small: an isolated number says little.

See also the public track record with every pick and the monthly results report.

To judge any tool, read the 7 questions to judge a betting AI.

And to understand the method, read our guide to AI for sports betting.

If you use the data in a story or a study, let us know.

It helps us improve the file.

Bet only what you can afford to lose.

Frequently asked questions

Where do I download Gambeta's picks record?

In two public files: one JSON and one CSV.

The addresses are in the table on this page.

No sign-up or payment is needed.

What license is it published under?

Under Creative Commons Attribution 4.0 (CC BY 4.0).

You can copy, analyze and reuse it, including commercially.

In return you must credit Gambeta and link to gambeta.ai.

How often is it updated?

It is updated every day.

Each pick enters before the match starts.

The result is filled in when the match ends.

How are results resolved?

Automatically, with data from independent public sources.

The sources are ESPN and TheSportsDB.

They are not edited by hand after the match.

Can I use it to train my own model?

Yes, the license allows it if you credit the source.

Keep in mind these are the picks our AI published, not every match.

So it is good for evaluation, not as a full match database.

How do I recalculate the yield?

Add up the ganancia column and divide it by the sum of stake.

Use the picks with a result of win, loss or void.

Multiplied by 100, the result is the yield as a percentage.