Skip to content
Trending storyDeveloping

GPT-6-Astra scores 98% on a physics research benchmark

1 report1 sourceupdated 2 hours ago

What happened

AI digest

On October 15, 2026, The Decoder reported on GPT-6-Astra's results on TheoPhysEval, a physics research benchmark. Tested in OpenAI's Codex environment with no human hints, the model solved 49 of 50 physics research problems, a 98% accuracy rate. On cost, the headline says about $2.40 per problem, while the article body says the full test cost roughly $116.

Written by AI from the coverage · updated 1 hour ago

Coverage

Follow the reports to see the story from different sides.

Oct 15
  1. The Decoder
    GPT-6-Astra solves 49 of 50 physics research problems at $2.40 apiece

    GPT-6-Astra 在 OpenAI 的 Codex 环境中,无人工提示地解出 TheoPhysEval 的 50 道物理研究题中的 49 道,正确率为 98%,整套测试花费约 $116。

Heat over time

Not enough continuous observations to draw a trend yet.