Seeing the same. Simulator will not run .load. FB22699606 filed. private func failingFunction() async throws { guard let embedding = NLContextualEmbedding(language: .english) else { return } print("hasAvailableAssets:", embedding.hasAvailableAssets) let result = try await embedding.requestAssets() print("requestAssets result:", result) try embedding.load() // <- throws here let r = try embedding.embeddingResult(for: "hello world", language: nil) print("length \(r.sequenceLength)") } Generates a 30s delay, then errors. hasAvailableAssets: false Type: stdio Failed to send CA Event for app launch measurements for ca_event_type: 0 event_name: com.apple.app_launch_measurement.FirstFramePresentationMetric Type: Error | Timestamp: 2026-05-04 19:39:07.618665-04:00 | Process: NLContextualEmbeddingLoadFailure | Library: libapp_launch_measurement.dylib | Subsystem: com.apple.app_launch_measurement | Category: General | TID: 0x225e218 Failed to send CA Event for app launch measurements for ca_event_type: 1 event_name: com.apple.app_launch_measurement.ExtendedLaunchMetrics Type: Error | Timestamp: 2026-05-04 19:39:07.727782-04:00 | Process: NLContextualEmbeddingLoadFailure | Library: libapp_launch_measurement.dylib | Subsystem: com.apple.app_launch_measurement | Category: General | TID: 0x225e21f objc[78708]: Class UIAccessibilityLoaderWebShared is implemented in both /Library/Developer/CoreSimulator/Volumes/iOS_23E244/Library/Developer/CoreSimulator/Profiles/Runtimes/iOS 26.4.simruntime/Contents/Resources/RuntimeRoot/System/Library/AccessibilityBundles/WebKit.axbundle/WebKit (0x109388398) and /Library/Developer/CoreSimulator/Volumes/iOS_23E244/Library/Developer/CoreSimulator/Profiles/Runtimes/iOS 26.4.simruntime/Contents/Resources/RuntimeRoot/System/Library/AccessibilityBundles/WebCore.axbundle/WebCore (0x117358310). This may cause spurious casting failures and mysterious crashes. One of the duplicates must be removed or renamed. Type: stdio filesystem error: in create_directories: Permission denied ["/var/db/com.apple.naturallanguaged/com.apple.e5rt.e5bundlecache"] Type: Error | Timestamp: 2026-05-04 19:39:37.695673-04:00 | Process: NLContextualEmbeddingLoadFailure | Library: NaturalLanguage | Subsystem: com.apple.NaturalLanguageFramework | Category: default | TID: 0x225e219 Compilation of '5C45D94E-BAB4-4927-94B6-8B5745C46289' model failed:Error Domain=NLNaturalLanguageErrorDomain Code=7 "Aborting repeated compilation request" UserInfo={NSLocalizedDescription=Aborting repeated compilation request} Caller bundle id: 'com.disparateelements.NLContextualEmbeddingLoadFailure' Type: Error | Timestamp: 2026-05-04 19:39:37.928174-04:00 | Process: NLContextualEmbeddingLoadFailure | Library: NaturalLanguage | Subsystem: com.apple.NaturalLanguageFramework | Category: default | TID: 0x225e219 Failed with error Aborting repeated compilation request Type: Error | Timestamp: 2026-05-04 19:39:37.928618-04:00 | Process: NLContextualEmbeddingLoadFailure | Library: NLContextualEmbeddingLoadFailure.debug.dylib | Subsystem: test | Category: NLContextualEmbeddingLoadFailure | TID: 0x225e091

I've heard from two affected customers that the problem has gone away. I'll be reaching out to other affected customers but I'm cautiously hopeful. I've been working furiously to set up a API proxy server to route calls from the app (thereby using the REST API on my server, and potentially getting rid of WeatherKit entirely). I still think I want to finish that work so I can have some insurance in case I need to suddenly switch weather providers. With a proxy server I can do that without a binary update on the App Store. The WeatherKit SDK is very convenient in how it doesn't require a proxy server to secure the API key, but this failure from the last couple of days has me spooked (especially given no acknowledgment from Apple that there was even a problem, and nothing ever showed up on their status board as far as I saw).

I may be seeing a similar issue. A few customer devices are getting this error on every WeatherKit request. I can't replicate it on my own devices, but I'm telling customers they should just request refunds because they can't get the service they are paying for. This is my post on the topic. https://developer.apple.com/forums/thread/789319 Did you ever find a solution for your dev device?

I'm also getting emails from users that my weatherkit-based app is failing to download data. I can run the same query on my own device and it works, so it must be a problem with their device in some way. My dev certificate was reissued about 3 weeks ago, but the affected users installed the app in the last couple of days, so I wouldn't think it should be a problem. Did you find a way to resolve this on your affected devices? I've asked my users to reboot their phones.

Thank you for your reply. I can't explain why you saw the unit test pass. I see passes in the simulator in Xcode 16 RC using iOS 17.4 and 17.5. It fails for me only in iOS 18 RC on the simulator. The full demo project is attached to the FB linked above if that is helpful. If I can provide any other diagnostics please let me know. We developed the app in iOS 17.0 and these unit tests passed all year in all point releases of iOS 17. We were also all green for later iOS 18 betas prior to the RC. I can't say for certain that there is no other impact to our app from this bug. We saw a change in concurrency behavior in iOS 18 RC compared to the betas (and compared to iOS 17.X) which revealed a race condition in our code. We had reliably "won" the race prior to the RC, but reliably lost in the RC. We had to refactor a chunk of code to get that to work, and we may have also fixed any issues coming from the ModelContext behavior above at the same time. The only app impact that seems like I can point specifically at this bug is in a view that took a ModelObject as an argument, rather than running its own @Query. We have two models @Model Parent { var children: [Child] } @Model Child {} The view below would continue to show the ContentUnavailableView even after our background process had updated the parent instance, and it's content was visible on other screens which were populated successfully with @Query variable. MyView { let parent: Parent // <- passed as argument, not query var body { if let children = parent.children, children.isEmpty == false { trendsContent } else { ContentUnavailableView(...) } } } The issue you linked where Query vars do not update when ModelActors update is one we saw during all early betas of iOS 18 (and iOS 17.0 betas as well), but have not been impacted by in any released iOS version.

It turned out that we could get this fixed pretty quickly in our app, but there was more to fix in our unit tests. A pattern we had adopted in unit tests was testing async operations performed by our ModelActors: Create a ModelContext in a unit test. Create a ModelObject instance in that unit test Context Send that ModelObject to a ModelActor to get updated Back in the unit test, confirm that the ModelObject was updated. It's that step 4 that was broken by this change in iOS 18 RC.

This bug is fixed in the iOS 18.1 build in Xcode 16.1 beta 1, iOS 18.1 (22B5023b). This iOS build also fixes the bug where ModelActor context saves were not triggering view updates.

You will need to handle this yourself in your code to remove the duplicate record. You can search for Core Data “deduplication” in your favorite search engine. I posted this thread asking for advice for how to solve this in Swift Data but at the time it was not possible. With the new iOS 18 history processing it may be possible, but I haven’t looked into it yet. That session mentions identifying changes that originate in other processes like widgets, but does not explicitly mention Cloud-originated changes. If anyone has tried this I’d like to hear if it works to identify CloudKit changes. If you are using Core Data there are good tools to process data changes that originate on another device. Apple’s best article on the topic is here and it should give you all of the code you need to solve this for a Core Data application. The short version is that in your example scenario you will see that both records are created, and both records will be synced to CloudKit, and both records will be sent to both devices. You can set up rules that will process incoming data that originated from a CloudKit sync, and identify if they are appropriate to preprocess in some way (such as by removing duplicates according to whatever business logic is appropriate for your app).

Unfortunately this bug is not fixed in Xcode 16 beta 5 / iOS 18.0 (22A5326g) beta. Correct me if I'm wrong, but doesn't this break the only "correct" way to do background work that appends new ModelObjects to an existing to-many relationship? Our app downloads data and creates new Items that populate an existing to-many relationship in a Location ModelObject. Our understanding is that the only idiomatic way to do this in the background with Swift Data is with a ModelActor, so this is how our app functions. It would be great for a DTS Engineer to confirm that what I describe above is the correct approach, and that the bug described in this post is expected to be fixed before iOS 18 RC. Otherwise our app requires some more significant rework to get ready for iOS 18, and it's not clear that there is another Swift-Data-intended way to accomplish this task.

Perhaps try defining a relationship in your models. https://developer.apple.com/documentation/swiftdata/defining-data-relationships-with-enumerations-and-model-classes/ Your current usage of these models is non-standard. I understand that this is a toy example, but it isn't normally done to create a Model inside another Model without defining a relationship between them. You would normally do something like this: import Foundation import SwiftData @Model final class Item { var firstArray: [String] var itemDetail: ItemDetail init(firstArray: [String], itemDetail: ItemDetail) { self.firstArray = firstArray self.itemDetail = itemDetail } } @Model class ItemDetail { var secondArray: [String] // VVV This is the new line that creates the relationship. @Relationship(inverse: \Item.itemDetail) var item: Item init(secondArray: [String]) { self.secondArray = secondArray } }

Hi @sendtobo I'm not seeing any change in behavior when I try your code. I added the following function to my example ModelActor: func createEmployeeThenAdd(CompanyWithID companyID: PersistentIdentifier?) { guard let companyID, let company: Company = self[companyID, as: Company.self] else { logger.error("Could not get a company") return } let newEmployee = Employee(company: nil) modelContext.insert(newEmployee) newEmployee.company = company logger.notice("Created employee \(newEmployee.name) in Company \(newEmployee.company?.name ?? "NIL")") try! modelContext.save() } When I run it from iOS 17, everything works normally. When I run it from iOS 18 beta 4, I see in the console the expected result with a relationship: Created employee 14E7161E in Company E3AD3F67 But the UI doesn't update (as you noted). However when I tap the "Add in View" button which runs on the main actor, I see the view update with both inserts (failed insert via ModelActor, successful via UI/main actor). Employee 14E7161E in company NIL Employee 81AA7153 in company E3AD3F67 I note in your example that you are adding companies.first. This means you are not getting the company to add in the same way as my example. I'm passing in the PersistentIdentifier and pulling the model object from the container with that reference, so my function doesn't have access to all companies. How are you getting the list of all companies to add one to the relationship in your example?

Updating because this bug is still present in Xcode 16 beta 4 in iOS 18 beta (22A5316f). I'm not aware of a workaround.

I updated FB14323984 with the details above. Thanks for your thoughts. I agree that given that the old date(from:) behavior was already used by devs in the field (including us!) that it should be maintained. What I should have said was that if a change in functionality was going to happen, then it should return nil, not January 1.

I agree that the test is odd. In our app we are generating a sequence of all of the start dates for weeks within a year by iterating over weekOfYear. Somehow we ended up with both year and yearForWeekOfYear specified (probably because year alone didn't give us the results we wanted). Dropping the year does indeed generate the same sequence of results as including it in both iOS 17 and 18 beta so it looks like we just make that change and are good. It seems to me that Calendar.date(from:) should return nil for cases where year and yearForWeekOfYear are specified, rather than January 1 of the specified year. Cheers!

The code you posted is generic to all Observation objects, including Models in SwiftData, so it is not going to be helpful. Your steps to reproduce are also not helping us to help you because of course we don't have your app's code. The most likely root cause of your crash is related to the fact that Views in iOS 18 are now implicitly annotated with @MainActor. If you are running code that requires it to be on the main actor, it will work silently in iOS 18, but will fail in earlier versions of iOS. Also, be aware that Apple will not approve any apps that use Xcode 16 until it reaches release candidate. You have to use Xcode 15 and target iOS 17 until then to be approvable. Good luck to you