GPT-6-Astra Scores 98% on Theoretical Physics Benchmark

GPT-6-Astra solved 49 out of 50 research-level theoretical physics problems on a recent benchmark test. The cost of running the model for the evaluation was reported as low. This result points to an expanded role for AI in scientific assistance, though independent validation of the findings has not yet been completed.

The benchmark’s design requires careful interpretation. Strong performance on a curated set of problems does not automatically translate to general scientific reasoning, and the specific conditions under which the model succeeded have not been fully detailed. Researchers and product teams in science, healthcare, and education are likely to weigh these results against the need for reproducible, externally verified outcomes before integrating such tools into critical workflows.

For professionals in creative, technical, and writing fields, the score signals a continuing shift in the capabilities of large language models applied to specialized domains. The test offers a data point on performance and cost, but its practical meaning will depend on further scrutiny of how the problems were selected and scored.