GPT-o4 is here – OpenAI’s comeback with powerful new models

OpenAI just released two impressive new AI models that could change how businesses approach their AI strategy. Let me break down what this means for you without the hype.

What’s new with OpenAI’s latest release

OpenAI has launched GPT-o3 and GPT-o4 Mini, their newest language models. The surprise here isn’t just their improved intelligence, but their native ability to use tools effectively right out of the gate.

According to OpenAI, these are their smartest models yet, representing a significant improvement in capabilities for everyone from casual users to advanced researchers.

What makes these models special?

Three key things stand out:

1. Built-in tool usage: Unlike previous models that required special integrations, these models can use tools like web browsing and code execution natively
2. Agentic behavior: They can chain multiple tools together to solve complex problems
3. Novel idea generation: OpenAI claims these are their first models capable of truly novel ideas

In a demo during their launch event, o3 analyzed a complex physics poster, recognized missing information, searched recent research papers, and calculated results that would have taken a human researcher days to produce.

Performance improvements are substantial

The benchmarks show impressive gains:

• On challenging math tests, o4 Mini achieves 92-93% accuracy (up from o1’s 74-79%)
• On code competitions, o3 and o4 Mini with terminal access rank in the top 200 coders worldwide
• Visual reasoning improved significantly, with o3 scoring 75% on challenging visual reasoning tests (up from 55% with o1)

Perhaps most interestingly,