Open almost any collections page in the Shopify Help Center right now and you'll hit the same note at the top:

A new collections model is replacing the legacy manual and smart collections model.

That isn't a rewording. On June 17, 2026, Shopify shipped GraphQL Admin API version 2026-07, which replaced a collection's single ruleSet with a multi-source model. The old split — a custom collection holding a hand-picked list, or a smart collection driven by rules — is gone. A collection can now be both at once.

If the new model has reached your admin, the Help Center says your existing collections were already converted automatically. You didn't do anything, and nothing asked you.

The change that actually costs money, though, isn't in the admin. It's in your installed apps. An app still calling a pre-2026-07 API doesn't see collections that use the new features — and it doesn't get an error either. Below: what became native, where the new ceilings are, and the two failure modes that are silent by design. Research date: September 1, 2026.

The short version

If you've been paying an app to nest collections, exclude products, or build size-specific collection pages, check your bill. All three are native now.

If you've been paying an app to reorder products by sales, stock, or views every morning, keep paying. Ordering logic is still not a native feature, and the new model didn't change that.

If you run a discount app, a cart transform, or anything built on Shopify Functions, verify 2026-07 support before you build your first variant-level collection. An app that only checks product-level membership will treat every variant in that collection as ineligible, with no error anywhere.

If you're shopping for a sorting app, compare billing denominators before features. Across the seven apps below, the meter runs on managed collections, sorts per month, product count, or monthly sessions. Two apps priced at $19 mean very different things.

What the new model actually changed

Shopify's migration guide is direct about the before-and-after. Before 2026-07, every collection was either a custom collection (a manual product list) or a smart collection (rules inside a single ruleSet). Mutations like collectionAddProducts only worked on custom collections. In 2026-07, those distinctions go away.

What replaced them is the source. A collection now holds one or more sources, and each source decides which products belong. The Help Center lists four kinds available from the admin:

• Products — include products from your catalog, automatically by condition or by hand
• Variants — include only specific variants
• Collection — pull in every product from another existing collection
• App — pull in a product list supplied by an app

The practical consequence: one collection can now hold automated conditions, hand-picked products, and exclusions at the same time. The old "you can't manually add a product to a smart collection" wall is down.

Nesting is native. The Help Center's example builds a Teaware collection out of existing Teapots, Steepers, and Infusers collections; when a source collection changes, the parent follows.

So are exclusions. You can exclude individual products or exclude an entire other collection, and the Help Center states that exclusions always override inclusions — whether the inclusion was automatic or manual.

Four ceilings you didn't have before

The new model came with store-wide caps. All four appear together on the Help Center's Creating collections and adding products page.

What's counted	Maximum per store
Collections containing any conditions	5,000
Collections containing variants (manual or automatic)	100
Collections that contain another collection	50
Collections that exclude another collection	5

Read the gap between those rows. You get 5,000 condition-driven collections, but only 50 that nest and only 5 that exclude a collection. A taxonomy where every parent category rolls up its children hits the wall at 50. And "exclude the clearance collection from this campaign collection" is a pattern you can use five times, store-wide.

Conditions have their own ceiling. The Understanding conditions for collections page states that a single collection can have a maximum of 60 total conditions. The legacy smart collections page carries the same 60 — and adds that a collection already over the limit can still be edited and have conditions deleted, but no new condition can be added.

One more gate worth knowing: the Starter plan doesn't support collections at all. The Help Center says you need Basic or higher.

Four condition behaviors Shopify documents but nobody expects

These are all in the Help Center. They're the reason a collection "looks broken" when it's actually working as specified.

Special characters in tags are collapsed. Set a condition for tag red-new and you also get products tagged red_new, red+new, and red&new. The Help Center says special characters are either ignored or treated as the same. If your tag scheme distinguishes sale-final from sale_final, it doesn't here.

There is no "greater than or equal to." Only is greater than and is less than exist. Shopify's own documented workaround is to offset by a cent: to catch products at $100 or less, write price is less than $100.01.

A compare-at price of 0 is not empty. Zero counts as a set value; the default is NULL, which compare-at conditions don't recognize at all. The Help Center actually suggests setting compare-at to $0.00 as the way to push a product out of a "compare-at price is set" collection.

One matching variant pulls in the whole product. On a Products source, a condition like inventory greater than 0 includes a product when any single variant matches — even if you selected "match all conditions." Five variants, four in stock, one sold out: the product is in. If you need variant precision, that's what the Variants source is for.

The Help Center also publishes a workaround for collections that contain the wrong products: duplicate an existing condition exactly, save, then delete the duplicate and save again, to force the collection to recalculate. Shopify documenting a manual cache-buster is itself a useful signal about how eventual this consistency is.

The two silent failures on the app side

Old API versions don't see new-model collections

The migration guide spells this out. In GraphQL Admin API versions before 2026-07, a collection using any new feature — multiple sources, exclusions, sub-collections, variant-targeted sources — behaves like this:

singular queries such as collection(id:) return null with an HTTP 200 response and no top-level GraphQL errors. Connection queries omit these collections. REST Admin API single-resource requests return 404 Not Found.

No error. No warning. The collection is simply absent. Every app that reads collections is exposed: product feeds, inventory sync, recommendation engines, sorting apps, ERP connectors.

Shopify's migration checklist asks developers to add a friendly fallback for collections that come back null, get omitted, or 404. Which tells you what happens in apps that haven't done it.

Discount logic returns false for every variant

Apps built on Shopify Functions — discounts, cart transforms, cart validation, delivery customization — have a sharper version of the same problem.

Product.inAnyCollection and Product.inCollections still work, but the migration guide notes they only check whether a product in its entirety belongs to the collection. Then:

On a variant-scoped collection, they return false for every variant.

Build a "size M only" collection, point a discount at it, and if the app still reads membership from Product, nothing is discounted. No error is raised. 2026-07 added inAnyCollection and inCollections to ProductVariant; the app has to have moved to them.

Ask your discount vendor about 2026-07 before you build a variant-level collection, not after.

Apps can now feed collections

The new model also opens a door. The App source in the admin lets an app supply the product list — the Help Center's examples are Complementary products and Trending products.

Developers get something further: an app can create a shareable conditions source and link it to many of a shop's collections at once. Deleting the shareable source automatically detaches it from every collection linked to it. One rule, maintained centrally, applied store-wide.

Seven sorting apps, four billing denominators

The Shopify App Store's Collections category lists 144 apps as of September 1, 2026. Ordering logic stayed out of the native model, so this category isn't going anywhere.

Here they are sorted by the thing that actually determines your bill. All prices are USD as listed on the App Store.

App	Developer	Rating (reviews)	Free plan	Paid entry	What the meter counts
ST: Product & Collection Sort	Green Apple Solutions Pvt Ltd	5.0 (236)	Yes	$19/mo	Sorts per month
KX AI Collections	Kimonix	5.0 (200)	No (14-day trial)	$19/mo	Managed collections
Sort'd ‑ Prime Collection Sort	Beefy Nachos Ltd	5.0 (134)	Yes	$29/mo	Monthly sessions
Jedi ‑ Collection Sort Pro	Jedi Apps	4.8 (113)	Yes	$2.99/mo	Managed collections
Bestsellers reSort ‑ Organizer	EGNITION	5.0 (71)	Free to install	$9.99/mo	Product count
Nada Hide Sold Out	Digismoothie	4.8 (23)	Dev stores only	$9.99/mo	Product count
SortWise Collection Sort	Mezereon Inc.	5.0 (22)	Yes	$9/mo	Collections + orders + products

Three things fall out of that table.

The two $19 tiers are not comparable. ST gives you 1,000 sorts a month with overage billing and no cap on how many collections you touch. KX AI Collections gives you 5 managed collections, one refresh per day, and a 1,000-order ceiling. A catalog with forty categories exhausts the second one immediately; KX's next tier is $79 for 50 collections, then $299 for 200.

ST has the most conditional free plan I found. Its free tier stacks three separate gates: 250 sorts per month, stores under 250 processed orders, and — unusually — "Shopify Basic/Professional plans only." Gating a free plan on the merchant's Shopify plan name didn't appear on any other listing I checked.

Sort'd is the only one metered on traffic. Premium covers 100,000 monthly sessions at $29; above that it's $70 to 1M, $200 to 5M, $350 to 10M. Two stores with identical revenue pay differently based on how much people browse.

On languages: of the seven, only Nada Hide Sold Out (Digismoothie) lists Japanese among its App Store languages, and even that listing carries the line that the developer doesn't offer direct support in Japanese. A translated interface and a support desk that speaks your language are separate purchases everywhere in this category.

Worth noting that presence in the category doesn't mean the app runs. Flexible Collection Sort (pluginappstore) sits at 2.3 stars across 10 reviews, with entries reading "They dont answer our mails. The app dont work anymore!!!!" from a store that had it installed for over three years.

Picking by situation

Nesting, exclusions, and variant collections cover you. Don't install anything. Just check your category design against the 50-nest, 5-exclude, and 100-variant-collection ceilings before you build it out.

You only want sold-out products pushed down. Bestsellers reSort is free to install up to 49 products and 99 collections. Nada is the only option here with a Japanese interface.

Many collections, re-sorted daily. Jedi ‑ Collection Sort Pro is $14/mo for 100 managed collections, the lowest per-collection rate in this set, and its free tier gives 10 collections with unlimited sorts. Its listing shows Claude MCP and Sidekick among its integrations, for driving sort rules from chat.

Large catalog, multi-signal merchandising. ST's reviews include an August 26, 2026 entry from a Shopify Plus store running roughly 2,000 collections and 100,000+ monthly orders, using multi-segment sorting to balance revenue, variant availability, and inventory.

Low traffic, but you want A/B testing. SortWise's free tier covers 10 collections, 500 monthly orders, and 10,000 products with hourly re-sorting and analytics. A/B testing starts at $9.

Before you install

• Does your category design fit inside 50 nested, 5 collection-level exclusions, and 100 variant-containing collections?
• Is any single collection approaching 60 conditions?
• Do your tags use symbols that collection conditions will collapse together?
• Has your discount or cart app confirmed 2026-07 support? (Ask before building a variant-level collection.)
• Are your feed and sync apps still returning every collection, or have some quietly gone empty?
• Does the sorting app's meter — collections, sorts, sessions, or products — match the axis your store grows along?
• Does the free plan gate on your Shopify plan or order volume?
• Are you treating "translated UI" and "support in your language" as separate requirements?

Wrapping up

Four steps, in order:

1. Check whether the new model has landed in your admin. If it has, your collections were converted for you, and nesting, exclusions, and variant-level collections are available.
2. If an app was providing any of those three, review the contract.
3. Confirm 2026-07 support with your discount, feed, and sync vendors before you build anything variant-scoped. Both failure modes here are silent.
4. Only if you need automated ordering, compare sorting apps — starting with the billing denominator, not the feature list.

App Store pricing and ratings change without notice. Everything above was verified on September 1, 2026; check the current listing before you commit.