Periodic says its lab-data model beats Astra on a science task
Periodic says it trained a trillion-parameter model on its own lab data that outperforms GPT-6 Astra on a critical scientific analysis task using relatively little compute.
Congrats @LiamFedus @ekindogus and the entire Periodic team! Amazing to see these results and even more amazing to see the lab they’ve built in person. Their thesis on real world lab data already seeing major impact: “Periodic Neon is an early example: using our lab data, we trained a trillion-parameter model that outperforms GPT-6 Astra on a critical scientific analysis task using relatively little compute. As we scale our labs and compute, we expect AI trained on our lab data to tackle increasingly complex scientific problems”
This post quotes another post: “OpenAI trains Neon materials model on 1,300 H200s, beating GPT-6 Astra on its own benchmark”