Hub Nexus
Updated

AuthorNo author yetClaim it

See something to improve? Propose a change.

Support

A chart illustrates the improvement of AI systems in performing longer software tasks over time. The horizontal axis spans from 2019 to mid 2025, marking the development of various AI models, such as GPT-2, GPT-3, GPT-3.5, and several iterations of GPT-4. The vertical axis indicates the length of time in minutes that tasks take human professionals. Key points highlighted include: - "GPT 3.5 (which came out in Spring 2022) could only do tasks that take humans a few seconds, such as selecting the right file" - "OpenAI's o3(which came out in April 2025) can do tasks on its own that take humans 20 minutes," such as finding and fixing small bugs in code - The observed trend shows a rapid progression in AI capability Accompanying notes indicate that the data is based on 170 tasks across fields like software engineering and machine learning. The source for this data is the Model Evaluation & Threat Research (METR) from 2025, presented under a Creative Commons attribution license.

How will artificial intelligence (AI) impact people’s jobs?

This question has no simple answer, but the more AI systems can independently carry out long, job-like tasks, the greater their impact will likely be.

The chart shows a trend in this direction for software-related tasks. The length of tasks — in terms of how long they take human professionals — that AIs can do on their own has increased quickly in the past couple of years.

Before 2023, even the best AI systems could only perform tasks that take people around 10 seconds, such as selecting the right file.

Today, the best AIs can fairly reliably (with an 80% success rate) do tasks that take people 20 minutes or more, such as finding and fixing bugs in code or configuring common software packages.

It’s unclear how much these results generalize; other factors, like reliability, need to be considered.

But AI capabilities continue to improve, and if developments keep pace for the next few years, we could see systems capable of performing tasks that take people days or even longer.

Read more about how we can help make our future with AI go well →

Where this page came from

This page was imported from Our World in Data. “The length of software tasks AI systems can do on their own has been increasing quickly” by Charlie Giattino, published by Our World in Data under CC BY 4.0. Changed here: set as a page, its interactive charts shown as pictures. Data from third parties keeps its own licence.

Nobody has written it yet — it is the source material at a new address, which is why search engines are asked to skip it and why no one earns from it. It is up for grabs: take it on, and it is yours to rewrite and to earn from.

LanguagesEnglish

Licence: CC BY 4.0 · Adapted from ourworldindata.org

1

0

0

0

Spinner Logo

Comments

Spinner Logo
Version: 2CC0 1.0 — public domain
The runaway star that left the Tarantula Nebula
Version: 2CC0 1.0 — public domain
The Blackwell School, where segregation had no law behind it
Version: 2CC0 1.0 — public domain
The Eagle Nebula, seen in the infrared
Version: 2CC0 1.0 — public domain
The house where the Equal Rights Amendment was written
Version: 2CC0 1.0 — public domain
The Aleutians, the forgotten front of the Second World War
Version: 2CC0 1.0 — public domain
The Cosmic Cliffs are not cliffs