Back to today’s briefing

Story perspectives

OpenAI's o3 Model Scores 25%, Sparking Benchmark Disputes

4/21/2025

50 7

Summary

Story summary

OpenAI's o3 AI model reports a 25% score on FrontierMath, contrasting with Epoch AI's independent benchmark of 10%.
Testing may have utilized a more advanced version than what is publicly available.
The public model is tailored for real-world applications, which could result in lower benchmark scores.
Benchmarking disputes are prevalent in the AI sector, highlighting concerns over transparency.

1 of 1

Story summary

OpenAI's o3 AI model reports a 25% score on FrontierMath, contrasting with Epoch AI's independent benchmark of 10%.
Testing may have utilized a more advanced version than what is publicly available.
The public model is tailored for real-world applications, which could result in lower benchmark scores.
Benchmarking disputes are prevalent in the AI sector, highlighting concerns over transparency.