How we test
Every result on this site says how we know it. If we have not confirmed something, we say that instead of guessing.
Which tools are compared
A tool belongs here if you can use it to produce a publication-grade 2D chemical drawing, which is what ChemDraw is for. Web editors built to be embedded in other software are compared separately. Viewers, converters and libraries such as Avogadro and Open Babel are listed with a reason, but not compared.
The drawings
Free tools are installed and run on a Mac (macOS 27, Apple M4). Browser tools are opened in Chromium. ChemDraw Prime 26.0 is included as the baseline, and because of its license we show only exported drawings of our test molecules, not its interface. Other paid tools are checked from their manuals.
Each tool gets the same three molecules, taken from PubChem. We use default settings and do not tidy the result. The one exception is ChemDraw, shown in its ACS Document 1996 style rather than its out-of-box style. SMILES is the first choice of input, because then the layout is the tool's own work. We paste the Kekulé SMILES that PubChem publishes, so ring drawing is also left to the tool. MOL is the fallback, used only when SMILES gives no drawing at all. A MOL file carries our own coordinates, so those drawings show the tool's style but not its layout, and the page says so. A SMILES drawing stays even if it overlaps, because that is what you would get.
| Molecule | Formula | Heavy atoms | Stereocenters | PubChem CID |
|---|---|---|---|---|
| caffeine | C8H10N4O2 | 14 | 0 | 2519 |
| paclitaxel | C47H51NO14 | 62 | 11 | 36314 |
| cholesterol | C27H46O | 28 | 8 | 5997 |
When a tool can write the structure back out, we compare its InChIKey with PubChem's. A match shows as "structure verified", which includes stereochemistry. "Stereochemistry changed" means the atoms and bonds survived but at least one stereocenter did not. Drawings with a problem stay on the page, so you can judge them yourself.
The feature checklist
The 65 features come from ReShiki's feature list, plus things only other tools do, such as TLC plates and peptide tools, so the list does not favor ReShiki. Each feature has a one-sentence test. Hands-on results take priority over manuals.
- Tested by running it
- Seen in the app
- The manual says yes
- Partly, with the reason shown
- Not supported
- Not established, which is not the same as no
Select a mark for its proof. A filled mark opens our screenshot, or the manual if the check left nothing to picture. A ring opens the manual.
"Not established" means we found nothing either way. A small project with a short README will show many of these, and that does not mean the feature is missing.
Sizes and speed
Download size is the installer we fetched. Installed size is the final folder on disk. For browser tools, the size is what the page transfers on a first visit.
Startup is the median of three launches in a row on the same Mac, timed until the first window appears. Memory is the actual memory in use, the macOS physical footprint shown in Activity Monitor, added up across the app's whole process tree. It is read after 10 quiet seconds on an empty document and again with paclitaxel loaded. For browser tools we also show the share used by an empty browser, so the two kinds of tool are not confused. For browser tools, startup is the median of three uncached loads. These timings show rough differences only: hosted tools also depend on the network and on their server.
Memory and startup were measured on the maintainer's own Mac (Apple M4, 32 GB, macOS 27.0.1, 2× display). They are not guaranteed values. Yours will differ with your hardware, macOS and app versions, window size and what else is running.
Memory depends on window size, because a window needs memory for every pixel it draws. ReShiki doubled when its window was enlarged from 1280 by 848 to 2144 by 1256 points, so we leave every app at its own default window size, on a 2× display.
Embedding
Embedding means another app or web page can place the working editor inside itself, using something the project offers for that purpose. Yes needs documentation. Partial means there is a real limit, such as a separately sold product, an undocumented route or a Java applet that browsers no longer run. No needs a source, such as a site that refuses to load in a frame and offers no other route. We fetched those response headers ourselves. Unknown means we could not confirm it either way. Office OLE is its own feature, so it does not count here.
Filling gaps from documentation
Where we could not test a feature, we read the tool's own documentation. A claim is kept only if its exact quote is on the cited page and an independent review agrees that the page really settles it. Most gaps stay Not established, because the documentation does not say either way.
Official demos
A video counts as official only if the tool's maker published it and its own website, documentation or repository links to it. We check that the link is really on that page and that the video still plays. If a page blocks our check, or a video cannot be embedded, we show a link and name the publisher where we can confirm it.
Listing order
Order is not a score. ChemDraw is first because it is the baseline. ReShiki is second because the maintainer of this site wrote it, and that position is a choice, not a measurement. Everything else is sorted by tools we ran first, then by features we tested by hand, then by features confirmed from manuals. This rewards what we could check, and it is still not a measure of quality. One exception: Octet Draw sits directly above ChemDraft, which is also a choice and not a measurement.
What we have not tested
47 of the 128 compared tools were not run: paid tools, tools for Windows or Linux only, and tools that would not start on our Mac. Their pages say which applies. Results apply to the exact version shown and may change with the next release.