Research · Updated 15 Oct, 08:54 pm IST
GPT-6-Astra solved 49 of 50 physics tasks at about $2.40 each
Why it matters for readers: Provides evidence that AI can perform many standard research tasks accurately and cheaply.
- GPT-6-Astra achieved correct results on 49 of 50 benchmark problems (98%).1
- The agent ran inside OpenAI's Codex coding environment during the test.1
- Average cost per problem was about $2.40, totaling roughly $116 for all 50 tasks.1
- The benchmark focused on 'turn-the-crank' routine research tasks across eleven subfields of theoretical physics.1
- Coordinators note these tasks are similar to initial research work often given to PhD students and exclude major open questions.1
Get a brief like this every morning
Uzha reads hundreds of sources and gives you the stories that matter for your work, with every source linked. Free.
Get started