Launch HN: Discovered Materials (YC P26) – AI agents to discover new materials
discoveredmaterials.com
August 12, 2026
8 min read
🔥🔥🔥🔥🔥
44/100
Summary
The Material Discovery Bench evaluates models on their ability to propose novel, BEOL-compatible crystalline materials that meet specific targets for thermal conductivity ($kappa), static dielectric constant ($epsilon), Young's modulus, and shear modulus. The benchmark utilizes a harness, tools, and graders to measure model performance based on these criteria.
Key Takeaways
- The benchmark aims to discover novel, BEOL-compatible crystalline materials that meet specific targets for thermal conductivity, static dielectric constant, Young's modulus, and shear modulus.
- Models are equipped with tools including web search capabilities, a coding sandbox, and machine learning tools to compute material properties such as dynamic stability and lattice thermal conductivity.
- The study utilizes machine learning interatomic potentials and various computational methods to evaluate the properties of proposed materials, including the use of the universal point edge transformer model.
- The model operates without a stopping condition, continuing until it encounters an error or exhausts its token budget of 100 million tokens.
Community Sentiment
Positives
- This project is the first to actually clarify which discovered materials are feasible, a significant leap forward in material discovery.
- Closing the computational-experimental loop is the main challenge, and there's real potential here — best of luck to the team!
- The approach of using LLMs for high-throughput material discovery shows promise, especially in less price-sensitive sectors like semiconductors.
Concerns
- Despite the hype around using AI for material discovery, previous attempts haven't made a real impact — skepticism is high.
- There's a concern that finding better materials is pointless if they are prohibitively expensive or hard to synthesize.