Last verified 19 August 2026.
Why this page exists
Most tools in this space treat their data sources as a competitive secret. Ours is the opposite position: the sources are public, free and better funded than anything we could build, so the honest thing is to name them and let you check.
There is a second reason, and it is the one that matters to an engineer. Coverage is not the same as conformance. A tool can cite every number perfectly and still not have the number you need. Knowing in advance which of those two situations you are in saves an afternoon.
The sources
Materials Project
Density-functional-theory calculations on inorganic crystalline materials, run at Lawrence Berkeley National Laboratory. Founded in 2011 with an initial emphasis on battery materials, and still the centre of gravity of the field.
| What it holds | Formation energies, crystal structures, band gaps, elastic tensors, phase stability, and battery-related properties for inorganic compounds |
| Scale | Over 130,000 inorganic compounds and around 35,000 molecules [figure needs a live check — the published summary carries no date] |
| Current release | v2026.04.13, live 8 June 2026 |
| Update cadence | Irregular. Releases are versioned by the date they become public; intervals have ranged from weeks to months |
| Licence | [VERIFY — still not retrievable. Rechecked 20 August 2026: the terms page returns 403 to automated retrieval. It has to be opened in a browser before this page is published] |
| One licence term we can source | Structures contributed from GNoME are licensed BY-NC — non-commercial — and explicitly accepting that licence is required to access the GNoME dataset. Stated on the Materials Project database-versions page, retrieved 20 August 2026 |
| Typical evidence level | E2 — computed and transparent, when quoted with its mp-id |
| Source | next-gen.materialsproject.org · database versions |
The GNoME term matters more than it looks. Materials Project is not one licence. A subset inside it carries a non-commercial condition, which is exactly the kind of clause that travels silently into a commercial product through an aggregator that flattened the provenance. It is also the reason the general licence has to be read in a browser rather than assumed: if part of the database is BY-NC, “Materials Project is open” is not a sentence worth relying on.
What it cannot tell you. It does not hold measured mechanical properties of commercial grades. There is no yield strength for 316L in it, because 316L is a specification, not a crystal structure.
OQMD — the Open Quantum Materials Database
DFT calculations from Northwestern University, notable for scale and for systematic coverage of hypothetical compounds rather than only known ones.
| What it holds | Formation energies, phase stability and convex-hull data for a very large set of compounds, including hypothetical ones |
| Scale | Passed one million compounds; documented by its maintainers in Reflections on one million compounds in the OQMD |
| Licence | [VERIFY — still not retrievable. Rechecked 20 August 2026: the download page is disallowed to automated retrieval, and on that date the site's robots.txt itself failed with a 502, so the request is refused rather than answered] |
| Typical evidence level | E2 |
| Source | oqmd.org · J. Phys. Mater., 10.1088/2515-7639/ac7ba9 |
What it cannot tell you. Coverage of hypothetical compounds is a strength for discovery and a hazard for selection: an entry existing does not mean the material has ever been made.
OMDB — the Organic Materials Database
Electronic structure data for three-dimensional organic crystals, developed and hosted at Nordita, the Nordic Institute for Theoretical Physics.
| What it holds | Band structures, densities of states and magnetic properties for organic crystals, metal-organic frameworks and molecular crystals |
| Licence | None stated. The site describes itself as open access, but its terms and conditions page covers user conduct and registration only and says nothing about reuse, redistribution or citation of the data. Checked 20 August 2026 |
| Typical evidence level | E2 |
| Source | omdb.mathub.io · terms · also reachable through OPTIMADE |
What it cannot tell you. “Organic crystals” is not the same as “polymers”. It does not cover engineering thermoplastics, elastomers or formulated systems.
And a note on the licence row. “None stated” is a finding, not a gap we failed to fill. The terms page exists, it is readable, and it does not license the data. Anyone redistributing OMDB values is doing so on an unstated basis, which is worth knowing before building on them.
OPTIMADE — the layer that connects them
Not a database. A common API specification that lets one query run against many structural databases at once, which is what makes it practical for a small team to read a large fraction of the open materials data landscape.
| Providers registered | 21 of 28 known providers, as of 19 August 2026 |
| Aggregate scale | Roughly 27.3 million structures across 45 aggregated sub-databases |
| Registered providers include | AFLOW, Alexandria, COD, JARVIS, Materials Project, MPDS, NOMAD, OMDB, OQMD, TCOD, 2DMatpedia |
| Source | optimade.org providers dashboard |
RepOD
Polish open research data repository, run on Dataverse by the Interdisciplinary Centre for Mathematical and Computational Modelling at the University of Warsaw. Used for locally deposited datasets where relevant.
| What it holds | General-purpose research data from Polish institutions — chemistry, physics, medicine and engineering are the largest categories |
| Scale | 4,346 datasets across 207 dataverses, 58,032 files, as displayed on the landing page on 20 August 2026 |
| Licence | Per dataset, not per repository. A Dataverse instance carries whatever licence each depositor chose, and the landing page states no default. There is no repository-wide licence to record, and quoting one would be wrong |
| Typical evidence level | Depends entirely on the deposit — a dataset with a method and a DOI can reach E4; one without either does not clear E1 |
| Source | repod.icm.edu.pl |
What it cannot tell you. Nothing systematically. It is a deposit repository, not a curated database, so coverage of any given material is accidental rather than designed.
The licence picture, which is worse than the field admits
Four databases, checked on 20 August 2026. Here is where the licences actually stand.
| Database | Licence status |
|---|---|
| Materials Project | Not readable. The terms page refuses automated retrieval. One subset inside it, GNoME, is separately non-commercial |
| OQMD | Not readable. Automated retrieval of the download page is disallowed |
| OMDB | None stated. The terms page exists and does not license the data |
| RepOD | Per dataset. No repository-wide licence exists to record |
Nothing here is an accusation. These are research projects, not publishers, and licensing is somebody’s third priority in every one of them. But it does mean that the sentence “open materials data” is doing more work than it can carry: two of these four do not tell you what you may do with the values, and one of the remaining two will not tell a machine at all.
That is why this page exists, and it is why we mark the rows we cannot resolve instead of writing “CC BY 4.0” from memory and moving on. If we cannot establish the licence, neither can you, and you are the one redistributing the number.
The same discipline applied to our own grade pages produces an audit of 191 values and 82 named gaps, none of them above E3. It is not a flattering page and it is the one we would point at first.
What we do not have, stated plainly
This is the part most vendors leave out, and it is the part you need before you sign up.
Commercial grade designations. You ask about 316L, S355, Inconel 718, Ti-6Al-4V.
The open databases are indexed by composition and crystal structure. The bridge between a trade
designation and a structure entry is incomplete on our side. The engineering-data layer that
closes it — Total Materia, MatWeb, Granta, MMPDS — is commercial, and we do not currently
license it. If you need a specification minimum for a named grade, we will show you what we
have, tell you it is not a specification minimum, and point you at the standard.
Formulated materials. Polymers, elastomers, composites, concretes, adhesives and coatings are defined by formulation and process history, not by composition alone. Our data layer does not represent them, so we do not answer questions about them with citations — and under our own framework, a value we cannot cite is a value we should not show.
Organic chemistry and life sciences. Coverage is effectively zero. Pharmaceutical packaging, excipients, extractables and leachables are outside our data layer.
Experimental crystal structures. ICSD and the Cambridge Structural Database are the authoritative sources and both are commercial. The open Crystallography Open Database is reachable through OPTIMADE and is not a substitute for either.
Where our data is strongest
Inorganic crystalline materials from computational databases: battery and energy-storage materials, catalysts, semiconductors, ceramics, and metallic systems treated at the composition level. This is where open DFT data is densest, most current and most frequently refreshed, and where a value from us most reliably reaches E2 with its entry ID attached.
How this connects to what you see in the product
The research agent carries this through to its output: every value shows the database it came from, the entry identifier, and — once the retrieval date ships — the date it was read. The evidence level is shown beside it, and the rules that govern what may be used for what are the same rules published in the open specification.
We do not currently meet our own Level 1 conformance. The gaps are listed at /trace/self-assessment, with dates.
What this page does not cover
- Data accuracy. Naming a source is not a warranty of the values in it. DFT results carry systematic errors that vary by property and by functional; the databases themselves document this and we do not restate it here.
- Licence compatibility for your use. Open licences carry attribution requirements. If you are redistributing values obtained through us, the upstream licence applies to you, not to us.
- Commercial databases. We do not have Total Materia, MatWeb, Granta or MMPDS, so we cannot tell you what is in them.
- Refresh guarantees. Upstream release cadences are irregular and outside our control. We record the date we read a value; we cannot promise how old it was when we read it.