Elements Claw - An AI agent for superconducting material discovery launched by Alibaba DAMO Academy

What is Elements Claw?

ElementsClaw is the industry's first AI agent for superconducting material discovery, launched by Alibaba DAMO Academy in collaboration with Renmin University of China and the University of Chinese Academy of Sciences. Employing a specialized-to-general fusion architecture, it combines the 1 billion-parameter atomic fundamental model Elements with a large-scale language model to automate the entire material screening process, from literature retrieval and property prediction to experimental design. The agent completed the screening of 2.4 million crystal structures in just 28 GPU hours, predicting 68,000 superconducting candidate materials and successfully experimentally validating four novel superconducting materials.

The main functions of Elements Claw

• Superconducting critical temperature predictionBased on the Elements-T model, the material Tc value is accurately predicted with an average error controlled within 1K.
• Superconductivity binary classificationThe AUC for determining whether a material has superconductivity using Elements-C is as high as 0.996.
• Thermodynamic stability assessment: Use Elements-E to evaluate the synthetic feasibility and structural stability of candidate materials.
• Crystal structure generationEnables de novo structural design through Elements-G, supporting motif guidance and exploration of entirely new phases.
• Intelligent literature synthesisLLM automatically retrieves, reads, and extracts scientific literature, uncovering scattered experimental knowledge.
• Experimental DesignIt combines numerical computation and semantic reasoning to output an executable synthesis path and verification strategy.

The technical principles of Elements Claw

• Atomic Fundamental Model ElementsBased on a 1 billion parameter model pre-trained with 125 million molecules and crystal structures, and employing the EquiformerV2 equivariant architecture, a unified representation of periodic crystals and aperiodic molecules is achieved. The dataset contains 85.1% of periodic crystals and 14.9% of aperiodic molecules, covering both equilibrium stable structures and high-energy unstable configurations, ensuring that the model learns robust cross-chemical domain representations.
• Multi-task pre-training strategyThe strategy involves performing coordinate/lattice parameter denoising on equilibrium inputs and simultaneously predicting system energy and atomic forces on non-equilibrium inputs. This approach enables the model to simultaneously grasp the static structural characteristics of stable matter and the dynamic force fields driving structural evolution.
• Private Network Convergence Agent ArchitectureThe proprietary model layer (Elements-T/C/E/G) is responsible for atomic-scale numerical computation; the general LLM layer is responsible for high-level semantic reasoning, literature synthesis, and decision planning. The two are coordinated through function calls, with the LLM orchestrating task chains and adaptively combining internal and external tools.
• Self-evolution mechanismAgent can extract new insights from literature and create new skill tools autonomously by fine-tuning Elements (such as Elements-C), thereby achieving a closed-loop capability expansion of "discovering needs - creating tools - verifying effects".

How to use Elements Claw

• Accessing Open DatabasesResearchers can log in to https://science.damo-academy.com to access the complete data of 2.4 million stable crystals predicted by ElementsClaw for free.
• Set exploration goalsEnter your research requirements into the system, such as "screening superconducting candidates with Tc greater than 4K in the Hf-Zr-Re ternary system".
• Start Smart FilteringElementsClaw automatically invokes Elements-T for high-throughput property prediction, while LLM retrieves and synthesizes relevant literature evidence.
• Get Candidate ListThe system outputs a list of high-confidence candidate materials, along with predicted Tc, stability indicators, synthesis feasibility, and literature support.
• Experimental synthesis verificationBased on the synthesis protocol recommended by the Agent, the target material was prepared in the laboratory and the resistance-temperature curves and magnetic susceptibility were tested.

Elements Claw's core advantages

• Efficiency RevolutionIt can complete the screening work that would take traditional methods decades in just 28 GPU hours, covering 2.4 million crystal structures.
• Leading in precisionThe superconductivity assessment AUC reached 0.996, and the Tc prediction MAE was only 0.992K, which is significantly better than the existing baseline.
• Self-evolutionNo new tools need to be developed manually; the agent can learn and create new skills autonomously from literature.
• Closed-loop verificationThe first batch of four AI-designed materials have all passed experimental verification, realizing the real-world application from prediction to synthesis.
• Open DataThe complete prediction database is made freely available to the global academic community to facilitate further research and collaboration.

Elements Claw project address

• Project official websitehttps://developer.damo-academy.com/material
• arXiv technical paper: https://arxiv.org/pdf/2604.23758

Comparison of Elements Claw with similar competing products

Application scenarios of Elements Claw

• Discovery of superconducting materials: Quickly identify high-critical-temperature superconducting candidates and significantly shorten the "trial and error" experimental cycle.
• Solid-state battery electrolyte developmentPredicting the stability and transport properties of ion conductor materials will accelerate the development of next-generation batteries.
• Heterogeneous catalyst designTo discover novel compounds with highly efficient catalytic active sites and promote chemical and energy conversion.
• Thermoelectric Materials ExplorationScreening for energy conversion materials that combine excellent Seebeck coefficient and low thermal conductivity.
• Database knowledge completionAutomatically retrieves known functional materials that have been missed by mainstream databases from scattered literature.