This week, in coordination with the broader Ember project, EmberData released 4.12, its final
4.x version, and began iterating towards the first release of the 5.x series.
A wrap on 4.x
This final release of the 4.x cycle introduced several new concepts to support a new paradigm for managing Requests and new capabilities for Caching Data.
The primary focus of these changes was to support caching Documents and to ensure that Resources are handled opaquely. Loosely defined: A Document encompasses the full response returned by an API request, and can typically be identified uniquely by url. Documents may contain many resources or references to resources, where a resource is defined as some data which your application or the API treats as uniquely identifiable.

Note: Historically, Model has mapped to a resource "type" and serves to present the data for a resource. It may be useful to think about resources as the backing data for a Model; however, resources can be far more generic than this and could even be anything. This includes raw strings, html, xml, buffers, images, blobs, and streams!

Collectively, these changes allow applications the capability of building highly advanced request and cache capabilities within familiar EmberData paradigms. We suggest reading the respective RFCs for RequestManager and Cache to gain a fuller appreciation of the motivations and capabilities they bring.
While these new primitives may not at first feel like a substantial shift, over the course of 5.x as the new request story is polished we expect for the recommended experience when using EmberData to significantly evolve, with that evolution driven by these changes.
The Polaris experience for EmberData will be focused around this new request experience, working with documents (and pagination) by default, and the new presentation class described below.
Looking ahead to 5.x
Throughout the 5.x cycle we expect to introduce two important new defaults while deprecating an old stalwart.
For nearly 17 years, Model has been a foundational primitive around which EmberData was understood. Since those earliest days, the language and ecosystem have evolved, the kinds of applications we build has evolved, and the patterns by which we access and mutate data have evolved. While Model has undergone small amounts of evolution in syntax, its underlying patterns have remained unevolved.
In more recent years, various 3rd party attempts have been made to address some of the shortcomings of the Model paradigm. ember-m3 explored what it might mean to have schema-driven models. ember-data-model-fragments offered a deep-tracking workaround. ember-data-storefront offered alternative data access patterns to simplify the mental model of asynchronous edges in relational data. ember-data-changetracker, ember-changeset, and ember-buffered-proxy offered mechanisms for streamlined mutation flows and more easily discardable changes.
We appreciate each of these addons for tackling challenging aspects of working with EmberData and addressing the genuine needs of our consumers, and we are thankful to all the maintainers over the years that helped these projects along.
In the absence the public APIs that EmberData now offers, many of these libraries built features on an unstable foundation and relied on private and intimate API contracts that were constantly changing. Consequently, applications using these addons have encountered significant difficulties when attempting to upgrade their version of EmberData.
So throughout the 3.x and 4.x cycles, we began evolving a new set of public APIs on top which we could support these ideas without compromise. RequestManager and Cache are the result of these efforts.
While this means these libraries can now rewrite to utilize public APIs and achieve improved stability, we also intend to bring most of the ideas explored by each of these libraries into the core recommended experience of EmberData itself.
What does this mean?
Today, Model serves two competing purposes: its static properties are used to declare schema information for attributes and relationships, while at runtime it is also the class we instantiate to present the data for a single resource out of the cache.
Importantly, both of these behaviors (schema and presentation) are just a configuration of public APIs (respectively, via registerSchema and the hook instantiateRecord) done for the @ember-data/model package. Schema no longer needs to be sourced from a Model, and record instances no longer need to be instances of Model. While this has been true for some time, when paired with the new request and cache APIs the potential now exists for far more declarative schema sources and far more powerful presentation classes.
Over the course of 5.x, we expect to introduce new defaults for declaring schemas, and new experiences for working with cached data: especially around presenting asynchronous data, paginated data, and handling mutation flows.
From a features perspective:

• an optional JS-like/Model-familiar schema syntax for build-time-compilation of schemas
  • by default these schemas would be imported as JSON by your application, though just-in-time fetching of schemas is also possible
• immutable record state
• deep tracking of dirty state on mutable copies of records
• paginated relationships by default
• always-sync access to current relationship state, with easy access request APIs that can be invoked from JS or templates. (This will be similar to the Document Presentation Class introduced in 4.12.

Lifting all Tides
A significant motivation for shipping Request and Cache in the 4.x series, even if it took till the end (as it did) is that we believe the easiest migration path for most applications looking to resolve deprecations around promise-proxies, async relationships, computed chains, and array-like APIs will be to incrementally migrate from Model to its replacement.
By shipping the request and cache APIs in 4.12, we've ensured that powerful replacements can be built that are compatible with 4.x enabling this migration path to be created. While we expect many applications might choose to implement their own presentation class – as often a custom class is capable of doing powerful things derived from Schema that a more generic presentation class could not do – we are additionally committing to making the official Model replacement compatible with EmberData 4.12
This means (for instance) that the migration path for an application using ember-data-model-fragments is to help that library become compatible with 4.12, update their application to 4.12, and begin incrementally migrating from Model+ModelFragments to the new presentation class which will offer deep-tracking out of the box.
For ember-data-model-fragments specifically, we commit to providing extensive time to assist in adding support for 4.12 to the current 3.28+ branch that some teams have been using today. In this way, we hope to give even those teams facing hurdles in the jump to 4.0 a carrot and a mechanism to leap quickly to 5.x.
EmberData 4.12 will become a Special LTS
In addition to the commitments above, we are planning on declaring EmberData 4.12 as a special LTS release.
4.12 will remain an actively supported LTS for the entire duration of the 5.x cycle, extending until the first 6.x LTS is released. This is in addition to our regular LTS support policy, and only applies to the EmberData project
This support extends beyond simple bug-fixes. If minor enhancements can be made to better support new presentation class implementations that support the 4.x series we will willingly accept them. Our goal is simple: we want no one left behind.