Story perspectives

OpenAI Launches FrontierScience Benchmark for AI Research Evaluation

12/16/2025

Summary

Story summary

OpenAI introduced FrontierScience, a benchmark to evaluate AI models' capabilities in scientific research.
The FrontierScience benchmark includes two tiers in physics, chemistry, and biology.
GPT-5.2 achieved 77.1% on the Olympiad tier and 25.3% on the Research tier.
Researchers say the benchmark advances the field but does not assess experimental execution or image analysis.
AI advances, particularly in reinforcement learning, are noted, yet reliable evaluation of specialized knowledge remains challenging.

1 of 1

Story summary

OpenAI introduced FrontierScience, a benchmark to evaluate AI models' capabilities in scientific research.
The FrontierScience benchmark includes two tiers in physics, chemistry, and biology.
GPT-5.2 achieved 77.1% on the Olympiad tier and 25.3% on the Research tier.
Researchers say the benchmark advances the field but does not assess experimental execution or image analysis.
AI advances, particularly in reinforcement learning, are noted, yet reliable evaluation of specialized knowledge remains challenging.