Periodic Labs' Neon tops GPT-6 Astra on materials analysis using just 1,300 H200s
LiamFedus · x · 2026-09-16
Liam Fedus' Periodic Labs built high-throughput materials labs in Menlo Park closing the experiment-model self-improvement loop. Its Neon model, mid-trained and RL'd on an open-source base with only 1,300 H200s plus months of lab data, beats GPT-6 Astra on their analysis benchmark.
- Method: splits discovery into choosing materials, synthesis planning, and interpreting results; models learn from past experiment records while labs run new ones
- Calibrated LLM judges turn scientific judgment into scalable feedback
- Focus: superconductors, magnets, and semiconductor materials, with real lab footage released
Related event: Periodic Labs Trains Materials Science Model Neon with Just 1,300 H200s(47 posts)→
More from Embodied
- Bionic Robobird Demonstrates Nature-Mimicking Flapping-Wing Flight — TinfoilTricorn · 2026-09-16
- Robotics researcher pushes back on 'omni embodiment' hype: it's the hands, not the abstraction — chris_j_paxton · 2026-09-16
- StarVLA's VLAct trains VLAs on 16 GPUs by reshaping action representations, not data scaling — jiqizhixin · 2026-09-16
- World Labs launches Atlas: an omni world model natively spanning text, images, video and 3D — YunzhuLiYZ · 2026-09-16
- Xynova's Flex 2 bionic hand weighs 400g with 23 DOF and lifts 12kg — creatoroff · 2026-09-16
- NVIDIA's Ming-Yu Liu Explains How Physical AI Learns Across Language, Video and Action — liu_mingyu · 2026-09-16