# Introduction
Humanity's Final Test (HLE) is a benchmark designed to measure the reasoning abilities and deep knowledge of many modern AI systems. Its defining characteristic: its basic assessment is taken to extremes. Think of it as an evolution of today's Turing tests, born a few decades ago.
This article goes into a bit of detail about this measure, explains why it was created, considers the various opinions from professional groups in the field about it, and concludes with a summary of the widely accepted decision.
# Why Was It Built, And What Does It Include?
Traditional testing methods used in older AI systems became obsolete as these systems evolved and began to obtain perfect scores without much effort. Because of this, the AI Security Center created a novel benchmark called HLE aside AI scale with the help of international experts. The benchmark is published on The environmentthe most prestigious scientific journal to date, January 2026. It is carefully designed to avoid repeating patterns as previous evaluation frameworks did.
So, what is HLE all about? Well, the test is to be taken by advanced AI systems such as language models, and it contains more than 2,500 expert-level questions covering more than 100 academic fields, including but not limited to physics, mathematics, biology, humanities, and many more. Importantly, the questions cannot be answered by memorization, and are not limited to simple information retrieval or multiple choice responses. Instead, they want complex thinking and deep understanding.
Here is an example of two such questions:
Two examples of HLE questions. Image source: AI Security Center
Let's talk about the results obtained by the most advanced models today: even the most sophisticated borderline models like GPT, Gemini, or Claude exceed the accuracy limit of 45-50%. The math speaks for itself on how difficult the test is. In addition, they often fail due to overconfident behavior in their wrongly answered questions.
# What Do Prominent Experts Think About HLE?
The honest answer is: there is little consensus on this. Opinion is divided among the technical, engineering, and academic communities, but there is a subtle, dominant tendency to accept something useful in HLE. There are important nuances, however.
In general, experts and a wide range of people familiar with HLE do not consider it a pointless initiative, but appeal to an exaggerated, seemingly marketing-oriented way of naming it.
To a large extent, there are three dominant groups of opinions about HLE:
// 1. HLE Is Really Useful And Necessary
About 60% of the opinions depend on this collective opinion, according to which there is a technical reason why HLE is important at the moment: past benchmarks and evaluation frameworks for AI systems, including the non-old language model such as Massive Multitask Language Understanding (MMLU), were saturated or outdated, almost all modern AI scored more than 90. This has made it difficult to really compare the latest models against each other to determine which one is the best. One important reason why HLE is praised by many experts is that it measures that AI is willing to say “I don't know” instead of fooling around with complex problems or questions it can't handle.
// 2. HLE is a distraction from Real AI
This skeptical opinion is accepted by about 30% of the opinions. These experts think that the test does not really test AI performance and effectiveness in everyday life situations, which is based entirely on superficial academic knowledge and hidden knowledge. Some developers even try to say, rather ironically, that as soon as AI starts to score more than 90% in HLE, enterprises will rush to build HLE 2, and so on, thus turning the marketing hamster wheel in favor of big companies.
// 3. HLE is flawless
This is the third and smallest of the three dominant views, and is discussed in data science forums, for example. They claim that HLE is flawed in some of the answers labeled as correct, especially in some niche questions from areas such as chemistry and advanced mathematics. Rather poetically, it was the most powerful AI systems themselves that first found such errors in the benchmark.
# Wrapping up
To summarize, the usefulness of HLE is not denied, and to some extent, its importance is emphasized by many experts, although its invention is widely considered as a mere marketing drama. Using this benchmark it seems very unlikely to determine the birth of great AI or its true evolution artificial general intelligence (AGI): a concept that has been talked about for years but is still more of a myth than a reality. However, benchmarking seems like a very ambitious tool to see which AI or company has the best model with memory and logical power.
Iván Palomares Carrascosa is a leader, author, speaker, and consultant in AI, machine learning, deep learning and LLMs. He trains and guides others in using AI in the real world.
