Neon: AI scientist trained on 1,300 H200s beats GPT-6 Astra on materials analysis benchmark
vwxyzjn · x · 2026-09-16
Liam Fedus's team unveiled Neon, an AI scientist for materials science, with real lab footage showing it hypothesizing crystal structures, reasoning about synthesis conditions, and iterating with tools until it finds physically realistic solutions for superconductor samples.
Key points:
- The team built high-throughput materials labs in Menlo Park creating a loop: labs generate fresh data, models learn from it, then suggest what to try next
- Using only 1,300 H200s plus months of experimental data for mid-training and RL, their open-source model surpassed GPT-6 Astra on their analysis benchmark
- Initial focus: hard materials science problems including superconductors, magnets, and semiconductor materials
- The authors argue more and more of science will look like this in coming years
More from Models
- LightOnOCR-3 Reads 16th-Century Manuscripts and Ancient Greek Without Training — IgorCarron · 2026-11-02
- 'You Can Feel the Gradient Slopes': Dev Contrasts Benchmaxxed RL Models vs General Ones — theteknosaur · 2026-11-02
- LightOnOCR-3 Launches: Tops FRBench, Adds Layout Structure and Chart Extraction — IgorCarron · 2026-11-02
- NVIDIA's Kumo Is a Family of Four Tabular Foundation Models, Not Just One — pandeyparul · 2026-11-02
- SemiAnalysis: safety classifiers are underrated — Meta's Muse tops cyber benchmark at 35.2% — ThomasScialom · 2026-11-02
- DeepSeek V4.1 Flash 1017 wins over a longtime closed-source-only user: fast, cheap, reliable — _bachrc · 2026-11-02