MaterialsMaterials Science

5 Best Materials Science AI Tools in 2026

Compare the top materials informatics platforms: Materials Project, AFLOW, NOMAD, GNoME, and SciRouter Crystal Explorer. Features, data, and accessibility ranked.

Ryan Bethencourt
April 16, 2026
11 min read

The Materials Informatics Landscape in 2026

Materials science is undergoing the same data-driven transformation that reshaped drug discovery, genomics, and climate modeling over the past decade. A new generation of AI tools makes it possible to predict material properties, screen candidate compositions, and discover novel structures without setting foot in a lab.

But the landscape is fragmented. Databases are scattered across different institutions with different APIs, data formats, and coverage. Some tools focus on data archiving, others on computation, and others on prediction. Choosing the right tool for your specific use case requires understanding what each one does well – and where it falls short.

Here are the five most impactful materials science AI tools available in 2026, ranked by practical utility for working researchers and developers.

1. Materials Project (Lawrence Berkeley National Laboratory)

The Materials Project is the most widely used computational materials database in the world. Founded in 2011 at Lawrence Berkeley National Laboratory, it hosts DFT-computed properties for over 150,000 inorganic compounds, including formation energies, band gaps, elastic properties, and electronic structures. Its companion library, pymatgen, is the de facto standard for computational materials science in Python.

What It Does Best

  • Comprehensive database of DFT-computed properties for known inorganic crystals
  • Phase diagram construction and thermodynamic stability analysis
  • Electronic band structure and density of states data
  • Well-documented Python API (mp-api) and web interface
  • Hosts GNoME data, making it a gateway to 2.2 million predicted structures

Limitations

  • DFT-only: no experimental data, ML predictions, or beyond-DFT calculations
  • Limited coverage of surfaces, interfaces, defects, and amorphous materials
  • API rate limits can be restrictive for large-scale screening workflows
  • Band gap values use PBE functional, which systematically underestimates gaps
Note
Materials Project is the best starting point for any materials informatics project. If the material you are interested in has been computed, you will likely find it here. Think of it as the PubChem of materials science – not exhaustive, but the first place you check.

2. AFLOW (Duke University)

AFLOW (Automatic FLOW for Materials Discovery) is both a database and a workflow engine. Developed at Duke University, it automates DFT calculations for crystalline materials using a standardized set of protocols. The AFLOW database contains over 3.5 million material property entries across more than 700,000 compounds, with particularly strong coverage of binary and ternary alloys and intermetallics.

What It Does Best

  • Automated, reproducible DFT workflows (AFLOW standardization ensures consistency)
  • Excellent coverage of metallic alloys, intermetallics, and Heusler compounds
  • AFLOW-ML: machine learning models for rapid property prediction
  • Symmetry analysis and prototype classification for crystal structures
  • RESTful API with comprehensive query capabilities

Limitations

  • Web interface is less intuitive than Materials Project
  • Documentation can be sparse for advanced API features
  • Focused primarily on bulk crystalline materials; limited surface or defect data
  • Slower adoption of modern ML integration compared to newer platforms

3. NOMAD (Novel Materials Discovery Laboratory)

NOMAD is the largest open archive of raw computational materials science data. Hosted by the European research consortium FAIR-DI, it stores the full input and output files from over 100 million DFT calculations, not just the extracted properties. This makes it invaluable for training ML models, reproducing calculations, and mining for patterns that extracted-property databases miss.

What It Does Best

  • Largest raw data archive: over 100 million calculations from dozens of DFT codes
  • FAIR data principles: findable, accessible, interoperable, reusable
  • Supports multiple DFT codes (VASP, Quantum ESPRESSO, FHI-aims, WIEN2k, and more)
  • Comprehensive metadata and provenance tracking
  • Encyclopedia interface for human-friendly property lookups

Limitations

  • Raw data focus means you often need to post-process results yourself
  • Data quality varies because it accepts community uploads without standardized workflows
  • API is powerful but has a steeper learning curve than Materials Project
  • Search and filtering for specific properties requires understanding the schema

4. SciRouter Crystal Explorer

SciRouter Crystal Explorer takes an API-first approach to materials science. Rather than being a database to query, it is a prediction engine: you provide a chemical composition and get back predicted properties, stability assessments, and simulated XRD patterns. The underlying models are trained on data from Materials Project, AFLOW, and NOMAD, combining the strengths of all three databases into a single, developer-friendly endpoint.

What It Does Best

  • API-first design: clean REST endpoints, JSON responses, standard authentication
  • Predicts properties for novel compositions not in any database
  • Batch screening: evaluate thousands of candidates in a single request
  • Hobbyist-friendly: free tier, no institutional affiliation required
  • Integrates with AI agents via MCP (Model Context Protocol) for autonomous workflows

Limitations

  • ML predictions are less accurate than full DFT calculations for edge cases
  • Newer platform with a smaller community than Materials Project
  • Does not store raw calculation data (not a database, but a prediction service)
  • Materials endpoints are still expanding; some property types are not yet available
Screen compositions with SciRouter Crystal Explorer
import requests

API_KEY = "sk-sci-your-api-key"
BASE = "https://api.scirouter.ai/v1"
HEADERS = {"Authorization": f"Bearer {API_KEY}"}

# Screen a batch of candidate compositions
response = requests.post(f"{BASE}/materials/properties",
    headers=HEADERS,
    json={
        "compositions": [
            "LiFePO4", "NaFePO4", "KFePO4",
            "LiMnPO4", "NaMnPO4", "LiCoPO4"
        ],
        "properties": ["formation_energy", "energy_above_hull",
                        "band_gap", "density"]
    })

results = response.json()["results"]
for comp, props in zip(response.json()["compositions"], results):
    stable = "STABLE" if props["energy_above_hull"] < 0.05 else "UNSTABLE"
    print(f"{comp:<12} E_hull={props['energy_above_hull']:.3f}  "
          f"Gap={props['band_gap']:.2f} eV  [{stable}]")

5. GNoME by Google DeepMind

GNoME (Graph Networks for Materials Exploration) is not a tool you run directly – it is a dataset and model that redefined the scale of computational materials discovery. Published in November 2023, GNoME used graph neural networks to predict 2.2 million new stable crystal structures, increasing the number of known stable inorganic crystals by an order of magnitude. The dataset has been integrated into the Materials Project, making it accessible through existing tools.

What It Does Best

  • Scale: 2.2 million predicted stable structures, 380,000 independently validated
  • Demonstrated that ML can systematically explore crystal chemistry at scale
  • Data is freely available through the Materials Project
  • Includes novel compositions for batteries, semiconductors, and superconductors
  • 736 predicted structures already synthesized and confirmed by robotic labs

Limitations

  • Not a tool you can run: it is a pre-computed dataset, not an interactive service
  • Predictions are for formation energy and stability only; other properties require separate calculations
  • No web interface or API of its own; access is through Materials Project
  • Quality varies: some predicted structures may not be synthesizable despite being thermodynamically stable
Tip
GNoME data is accessible through the Materials Project API. When querying Materials Project, you can filter for GNoME-predicted structures specifically, or include them alongside traditionally computed entries. This makes GNoME the largest expansion of accessible materials data in the history of the field.

Head-to-Head Comparison

Choosing the right tool depends on your use case. Here is how the five platforms compare across key dimensions:

  • For property lookups of known materials: Materials Project is the best starting point. Its database is comprehensive, well-documented, and easy to query.
  • For alloy and intermetallic research: AFLOW has the strongest coverage of metallic systems and automated workflow tools.
  • For training ML models: NOMAD provides the raw calculation data that pre-processed databases strip away. If you need input files, convergence data, or intermediate results, NOMAD is the source.
  • For screening novel compositions: SciRouter Crystal Explorer predicts properties for any composition, including ones absent from all databases. It is the only tool that works as a prediction engine rather than a lookup service.
  • For exploring newly discovered crystals: GNoME (via Materials Project) gives you access to 2.2 million predicted structures that were not in any database before 2023.

Building a Multi-Tool Workflow

In practice, the most effective approach combines multiple tools. A realistic materials discovery workflow might look like this:

  • Step 1 – Literature and database search: Query Materials Project for known compounds in your target composition space. Check what has already been computed and characterized.
  • Step 2 – Expand the search: Use SciRouter to predict properties for novel compositions that are not in any database. Screen thousands of candidates in minutes.
  • Step 3 – Validate promising hits: Cross-reference top candidates against GNoME predictions for stability. Check AFLOW for related alloy systems.
  • Step 4 – Deep analysis: For the best candidates, pull raw DFT data from NOMAD or run new calculations to validate ML predictions with full quantum mechanical accuracy.
  • Step 5 – Synthesize: Send your top 10–20 candidates to the lab for experimental validation.

Next Steps

To dive deeper into the methods and applications behind these tools:

Ready to start screening materials? Open the Crystal Explorer Studio or get a free API key to query materials properties programmatically.

Frequently Asked Questions

Which materials science AI tool should I start with?

Start with the Materials Project if you need to look up known materials and their DFT-computed properties. It has the most comprehensive database and the best documentation. If you need programmatic API access for screening or integration into automated workflows, SciRouter Crystal Explorer provides a developer-friendly REST API. If you are looking for raw computational data to train your own ML models, NOMAD has the largest collection of DFT calculations.

Are these tools free to use?

Materials Project, AFLOW, and NOMAD are free and publicly funded by government research agencies. GNoME data is freely available through the Materials Project. SciRouter offers a free tier with 500 API calls per month, with paid plans for higher throughput. All five tools provide open access to materials data, though API rate limits and data download restrictions vary.

Can I use these tools for commercial research?

Yes, all five tools allow commercial use of their data and APIs. Materials Project data is released under a Creative Commons license. AFLOW and NOMAD data are similarly open. SciRouter API access is governed by its terms of service, which permit commercial use. However, always check the specific licensing terms for derived datasets or models, as some restrictions may apply to redistributed data.

How do I choose between Materials Project and AFLOW?

Materials Project has a larger database of unique materials (over 150,000 compounds), better documentation, and a more modern web interface. AFLOW has stronger automated workflow tools and better coverage of binary and ternary alloys. If you are doing one-off property lookups, Materials Project is more user-friendly. If you are running systematic studies across composition spaces, AFLOW workflows may be more efficient.

What programming skills do I need to use these tools?

Basic Python is sufficient for all five tools. Materials Project has an official Python client (pymatgen + mp-api). AFLOW has the aflow Python library. NOMAD has a REST API accessible with requests. SciRouter provides a REST API that works with any HTTP client. GNoME data can be accessed through the Materials Project API. No specialized materials science coding skills are required beyond calling APIs and parsing JSON responses.

Can these tools predict properties for materials that do not exist yet?

Materials Project, AFLOW, and NOMAD are primarily databases of computed properties for known or hypothesized structures. They can tell you about materials that have been calculated but not necessarily synthesized. GNoME extends this by predicting stability for 2.2 million new compositions. SciRouter ML models can predict properties for any arbitrary composition, including those never seen in any database, making it the most flexible for novel materials exploration.

Try this yourself

500 free credits. No credit card required.