The new models from OpenAI o3 and o3-mini can do what only the human mind has done so far

OpenAI wrapped up its 12-day “Shipmas” series with a big announcement – the launch of the model o3the successor to the o1, along with its lightweight version o3-mini. These models bring significant improvements in reasoning, coding, and solving complex problems. They are said to be able to do what only humans have been able to do so far and will be much more intelligent. The news was presented by the company itself video.

What is o3 and o3-mini?

Model o3 it is designed to be able to analyze complex tasks and verify its own answers. Its smaller “brother” o3-mini is a simplified version, optimized for specific tasks with lower performance requirements.

Unlike the previous model o1, the new generation allows setting the level of “thinking” (low, medium or high computing capacity). The higher the level, the more accurate and reliable the model’s answers.

AGI: Reality or audacious goal?

OpenAI has indicated that o3 approaches k under certain conditions artificial general intelligence (AGI). It is defined as a system capable of surpassing human abilities in economically significant work. According to the ARC-AGI benchmark, o3 scored 87.5% on high compute settings, significantly outperforming o1. Even at the lowest computing setting, it achieved three times better results.

o3 models use the “chain of thought” technology, which allows them to gradually analyze the task, plan steps and summarize the best solutions. This ability brings benefits in physics, math and science. However, the process is more time-consuming – solutions can take seconds to minutes.

Although the o3 is more reliable and accurate, it is still not flawless. Models can be wrong and occasionally make basic mistakes, for example in games like tic-tac-toe.

Who are they intended for?

The o3 model is suitable for organizations and researchers who need high performance to solve complex tasks in areas such as coding, mathematics and science. The lower model o3-mini is ideal for developers and companies looking for flexible and cost-effective AI solutions with the ability to adapt the level of reasoning according to specific needs.

Availability

Both models are currently undergoing internal testing. OpenAI plans to release o3-mini to the public at the end of January 2025 and the full version of o3 shortly thereafter. Until January 10, 2025, researchers can request early access for security testing purposes.

It is highly unlikely that the new models will be available for free. OpenAI usually offers limited free access to basic models, while advanced versions tend to be part of paid services.

You might be interested in:

The question of why OpenAI skipped the “o2” name has a simple answer. The company wanted to avoid a conflict with the British telecommunications operator O2. The CEO of OpenAI, Sam Altman, confirmed this during the livestream.

You can also discuss this article on our website AndroidForum. If you’re looking for an answer to a problem with your mobile device, app or accessory, ask in our new AndroidFore.