AnyLogic 9 Technology Preview Review: The Future Is Taking Shape!
- Jaco-Ben Vosloo
- 6 hours ago
- 12 min read
The past year or two has been exceptionally busy with project work, leaving me with very little time to write for the blog. Now that I am returning to it, I thought the best place to restart would be with something timely: my recent testing and review of AnyLogic 9.
In this post, I share what impressed me, where AnyLogic 8 still has the advantage, and what the new version may tell us about the future of simulation modeling. And, if someone from the AnyLogic development team happens to be reading, perhaps one or two of my findings might even nudge a feature slightly higher up the priority list. 😉
I hope this marks the beginning of more frequent blog posts again.
Disclaimer: These are my personal views, based on my knowledge and experience at the time of writing. AnyLogic 9 is still under active development, so some, most likely all, of these opinions may change as the product continues to evolve.
Overview
AnyLogic is moving from a mature desktop application toward a browser-based, cloud-connected, fully integrated simulation environment!
AnyLogic 9 has been “coming soon” for a long time, ever since it was unveiled at the AnyLogic conference in 2019 as an R&D exercise in the AnyLogic and AnyLogic Cloud roadmap. But it seems, like all good things in life, it takes time and significant effort to complete.
It is not simply AnyLogic 8 with a fresh toolbar. It is also not yet a finished successor that every modeler should use for every project. It is the start of a much larger change: moving model development into a browser and connecting development, storage, execution, experiments, dashboards and collaboration through AnyLogic Cloud.
That is a more ambitious goal than a normal desktop-software upgrade.
After reviewing the current preview, my conclusion is deliberately mixed. AnyLogic 9 shows a convincing direction and already contains several thoughtful improvements. It is easier to access, its interface feels more modern, and its integration with the Cloud could remove a great deal of friction from the simulation lifecycle. At the same time, AnyLogic 8 remains the mature production tool, continues to receive important new features, and still offers capabilities that experienced users depend on.
My short verdict is this:
Try AnyLogic 9 now to understand where the platform is going. Keep AnyLogic 8 for production work until the specific features, libraries and workflows your model needs have been proven in version 9. (For me these are things like Git integration and other niche features like Project Lombok ;-) )
If you want to do your own review of AnyLogic 9 follow the official blog here.
What I reviewed
This article reviews the current AnyLogic 9 Technology Preview as a modeler, not as a final-product benchmark.
Version note: This review was checked against AnyLogic 9 Technology Preview 0.5 and the official release information available on 31 July 2026. AnyLogic 9 is moving quickly, so individual limitations may change after publication.
I focused on 3 questions:
What is genuinely improving compared with AnyLogic 8?
Where does AnyLogic 8 remain stronger? (For now)
Does a visual, drag-and-drop modeling environment still make sense in an AI-first future?
...That last question may turn out to be the most important one.
What already works better in AnyLogic 9
1. Browser-based modeling lowers the barrier to entry
Opening a browser is easier than managing an installed development environment. It reduces operating-system differences, makes it easier to move between computers, and gives new users a direct path from a shared Cloud model to its source.
For teaching, workshops and collaborative reviews, that is powerful. A student or stakeholder does not first need to understand installations, workspaces and local file paths before seeing a model. A modeler can move from the model repository into the editor and then into an experiment without repeatedly changing tools.
It also makes updates less disruptive in the SaaS version. The provider manages the editor and infrastructure instead of every modeler maintaining a separate installation.
This convenience is not free. Browser access and cloud dependency create their own questions, but it is a real improvement for accessibility.
2. The interface has been reconsidered rather than merely repainted
The workspace, project view, palette view, toolbar and properties have been redesigned. One small but meaningful example is that some space-markup geometry can be drawn directly from its properties.
The flowchart editor has also been a focus since the early previews. AnyLogic demonstrated smarter connectors, suggested downstream blocks, automatic connector reshaping and alignment guides. These are not revolutionary features in isolation, but they reduce the tiny interruptions that accumulate during a long modeling session.
That is the right place to improve a visual modeling tool. The value of drag-and-drop modeling does not come from the ability to draw a rectangle. It comes from expressing simulation concepts, queues, resources, delays, networks and states, without rebuilding those concepts from first principles.
3. Cloud integration is part of the workflow, not an export destination
In AnyLogic 8, the Cloud often feels like somewhere a model goes after development. In AnyLogic 9, it is the starting point.
This offers a much more direct route between:
model versions;
the model editor;
experiments and dashboards;
completed runs;
scenario comparison;
sharing and access control;
APIs and external applications.
The benefit is not merely fewer button clicks. It could make model governance more visible. Versions, experiments and results are less likely to become unrelated folders on one modeler's machine.
For operational models and digital twins that are updated and reused, this is more important than a prettier toolbar.
My Take:
However, this is also where one of my biggest reservations begins. I am a big fan of Git and of the way it enables concurrent model development. On a real project, we may have one modeler working on the user interface while another develops the logic of a particular agent and someone else works on data integration or experiments. They can work in parallel, review one another's changes and merge those changes into a testable version instead of taking turns editing the entire model.
The multipart ALPX format introduced in AnyLogic 8.9 made that workflow much more practical. Splitting agents, functions, classes, presentation elements and other model components across a structured set of files gives Git smaller and more meaningful changes to compare. It is not immune to merge conflicts, especially when two people edit the same model element, but it has significantly improved our ability to develop concurrently. In our team, that shortens the time between working versions that can be tested, analyzed or delivered to a client.
AnyLogic Cloud's model versions are useful, but they are not the same thing as source control. They preserve distinct model states and their experiments, and a version can be duplicated, downloaded or opened for editing. At the time of writing, however, the public documentation does not describe Git integration or an equivalent branch-and-merge workflow in the AnyLogic 9 Cloud editor.
That distinction is crucial. A sequence of saved versions provides history; Git enables several people to change the model at the same time and then combine their work. Without that capability, development must become more serial, or the team must create a manual process for coordinating and reintegrating changes.
For us, this is a major stumbling block to moving real projects completely into AnyLogic 9 and the Cloud. The integrated lifecycle is attractive, but until the web-based workflow supports Git, or offers collaboration with equivalent branching, comparison and merging, AnyLogic 8.9 with ALPX and Git remains the stronger environment for serious multi-developer projects.
4. The Java foundation is being preserved
AnyLogic's graphical interface has always been only one layer of the product. Underneath it is Java, and that remains one of the platform's greatest strengths.
The current AnyLogic 9 release notes state that Java classes and interfaces are fully supported, and that third-party JAR files can be added as model dependencies. This is essential for larger models. Real projects need domain classes, algorithms, reusable utilities, data connectors and testable code (not only palette blocks!).
It also means that the move to the web editor is not a move to a closed, low-code toy. The path from visual model to serious software engineering remains open.
Note: I am yet to figure out how to get my integrated Unit testing using JUNIT 5 to work in AnyLogic 9... see here for an example I presented at the AnyLogic conference in 2024.
This Java foundation and ability to create low-level code is also key in the age of AI (See the end of this blog post)
5. Important modeling and presentation capabilities are arriving steadily
Technology Preview 0.5 added shapefile import and conversion to space markup, editable cells in the built-in database, skyboxes in 3D windows, poses for 3D figures and combo-box support for parameters with arbitrary types. Earlier 2026 previews added full statechart-palette support, third-party database connectivity, random-seed settings, database table creation and deletion, and improvements to charts and 3D objects.
These additions are significant because they cover different parts of a real project: model logic, reproducibility, data, spatial modeling, controls and communication.
The pace also matters. Five preview releases appeared between March and June 2026. This is not an abandoned prototype; it is an actively advancing product.
Where AnyLogic 8 is still better (for now)
The strongest criticism of AnyLogic 9 is also the fairest: it is a technology preview.
Comparing it with AnyLogic 8 is not comparing two equally mature releases. AnyLogic 8 has years of production use, a broad modeling surface and a very recent release history of its own. AnyLogic 8.9.9, released in June 2026, added multiload transporters, tugger trains and built-in stacker cranes for automated storage and retrieval systems.
Version 8 is not standing still while version 9 catches up.
1. Feature completeness still favors AnyLogic 8
The current AnyLogic 9 still has System Dynamics under development and some internal work remaining on Material Handling Library blocks and markup.
The company's 2025 roadmap also placed several familiar capabilities on the journey toward parity rather than in the completed column.
The precise gap changes with every preview. That is why I would not publish a static “unsupported features” list and assume it will remain correct. I would test the actual model requirements instead.
2. AnyLogic 8 has the mature power-user workflow
Recent AnyLogic 8 releases include built-in Git support, Java 17, the multipart ALPX format, an integrated function profiler, favorites, folders for agents and Java classes, a chart-creation wizard, reusable element templates, a live 3D preview, Python-driven design-time model generation and continuing improvements across the specialist libraries.
Those features matter to experienced modelers because productivity is more than drawing a first flowchart. It includes navigating a large model, debugging generated and custom code, reviewing changes, profiling slow functions, reusing standard configurations and resolving mistakes quickly.
AnyLogic 9 has modernized the visible surface. It still has to prove that the complete daily workflow is at least as capable as the mature desktop environment.
3. A cloud-first editor changes the risk profile
The browser model is convenient, but some projects have requirements that convenience does not answer:
Can development continue during an internet outage?
Where is model source stored, and under which jurisdiction?
Can a client prohibit source code or data from entering a public SaaS environment?
How are backups, retention, access removal and audit logs handled?
Can the complete toolchain run inside a private network?
What happens if a long editing session loses connectivity?
AnyLogic has discussed SaaS, server-based and desktop stages for version 9, and AnyLogic Private Cloud addresses many enterprise deployment needs. Nevertheless, a modeler should evaluate the edition that exists for their use case, not a future deployment option on a roadmap.
For sensitive client work, architecture and governance are product features.
Is AnyLogic 8/9 ready for AI-first modeling?
AI does not sit outside the transition from AnyLogic 8.9 to AnyLogic 9. It can already assist through Java and the structured ALPX format, while AnyLogic 9 creates the opportunity for AI to become part of the modeling environment itself.
AnyLogic was one of the first, and in my mind best, tools to make complex simulation more accessible to ordinary humans.
That accessibility became a competitive advantage. A modeler could drag a Source, Queue, Service and Sink onto a canvas, connect them, add a resource pool, place buttons and charts on a presentation and create a working animated model without writing an entire simulation engine or user interface from scratch.
The libraries gave us ready-made objects. The palettes made them discoverable. The properties exposed common decisions. The animation made the model easier to explain. AnyLogic did a remarkable job of making sophisticated simulation visual and approachable.
But AI changes the economics of interfaces.
A human-friendly interface can be an AI-hostile interface
A graphical interface is easy for a person to explore. It is much harder for an AI agent to manipulate reliably.
To add a button through a visual IDE, an agent may need to:
find the correct palette;
drag the correct object to the correct canvas;
place and resize it;
select it;
find the correct property;
insert an action;
recognize whether the interface changed;
recover if a dialog, selection or drag operation behaved differently.
In a conventional Java project, an AI can often create the equivalent interface by editing text. It has an enormous public ecosystem of Java examples, APIs, tests and error messages to learn from. Text is searchable, diffable and reproducible. A drag operation is none of those things unless the application exposes it as a stable command or API.
This creates an uncomfortable possibility: the visual interface that made AnyLogic easier for humans could become a bottleneck when AI starts doing more of the building.
That does not mean visual modeling is obsolete. It means the visual interface cannot remain the only effective interface.
AnyLogic's Java foundation may become its AI advantage!
Fortunately, AnyLogic has an unusually strong answer to this problem: it is Java-based.
AI assistants can already help with the parts of an AnyLogic model that are expressed as code:
complex functions;
Java classes and interfaces;
data transformations;
collections and algorithms;
statechart actions;
database queries;
validation checks;
refactoring and documentation;
unit-like test utilities.
The multipart ALPX model format, introduced in AnyLogic 8.9, also helps. Instead of keeping almost everything in one large model file, it separates agents, Java functions, classes, connectors, variables, presentations and experiments into a structured collection of files.
In my experience, tools such as GitHub Copilot and AI agents working through VS Code can understand useful parts of this structure. They can often help edit functions, variables and Java classes, and can sometimes make controlled changes to the model definition.
They do not understand every AnyLogic relationship perfectly, and any generated change must be opened, compiled and checked in AnyLogic, but the format gives them something far more useful than pixels on a screen. Read my blog about how you can already use AI to enhance your AnyLogic modeling here or watch the YouTube video here.
This is why the Java and ALPX foundations are not implementation trivia. They are strategic assets.
The real solution is to make visual modeling programmable
The best future is not a contest between “drag and drop” and “pure code.” It is a model that can be manipulated safely through both.
AnyLogic 8.9.7 already introduced a design-time Python API that can generate elements in a model's graphical and logical structure. That is exactly the kind of bridge an AI agent needs: a documented way to say “create this agent, add these parameters, build this flowchart and place this markup” without pretending to be a human hand operating a mouse.
More importantly, AnyLogic is not ignoring the issue. At the 2025 AnyLogic conference, the company showed research into AI-assisted simulation co-design. You can review the published presentation and see how AI is helping with conceptualization, logic, flowcharts, statecharts, space markup, animation, debugging, experiment planning, dashboards, output interpretation and model improvement.
The same roadmap imagines three stages:
AI as a separate helper;
AI built into AnyLogic for specific tasks;
AI as the primary interface, orchestrating AnyLogic and other tools behind the scenes.
That last stage directly addresses the weakness of the graphical interface. An integrated agent does not need to guess where a button is on the screen if the IDE can give it the model structure, the library definitions and a safe set of model-editing operations.
The visual model still has an important future!
During the Winter Simulation Conference in December 2025, the AnyLogic team demoed a preview of how AI can generate a basic model from scratch as part of the UI and normal model building workflow!
AI may reduce the amount of manual dragging and dropping, but it does not remove the value of a visual model.
Visuals help humans:
inspect structure;
spot missing paths and strange connections;
communicate logic to subject-matter experts;
review an AI-generated model and provide it the feedback it needs to improve the model;
understand behavior during animation;
challenge assumptions;
build trust.
The future visual IDE may therefore be less important as the only place where every object is manually created, and more important as a place where people review, guide and validate what humans and agents create together.
In other words, AI should not make AnyLogic less visual. It should make the visual layer less of an automation barrier.
That is my assessment of AnyLogic 9 today and my view of where AI-assisted simulation could go next.
As always, these opinions are entirely my own, based on my knowledge and experience at the time of writing, and they will almost certainly evolve as AnyLogic 9 continues to develop.
If you disagree, have had a different experience, or simply want to continue the discussion, please feel free to contact me on LinkedIn or by email. I would genuinely enjoy hearing what other modelers think. Whatever the details of the eventual destination, I am excited to see what AnyLogic 9, AI and the next generation of simulation modeling have in store for us.
What next?
If you liked this post, you could read more posts by following the links at the top of the page. Why not subscribe to our blog or follow us on any of the social media accounts for future updates? You can also join the mobile app here!
If you really want to make a difference in supporting us, please consider joining our Patreon community here
If you want to contact us for some advice, maybe a potential partnership or project or just to say "Hi!", feel free to get in touch here, and we will get back to you soon!
Comments