Unity AI Beta 2026 is here!

Hello everyone,

First, thank you to each of you who has spent time testing our AI features. We truly appreciate the effort and feedback you’ve shared, and we hope the initial beta has delivered value since its release last year. To anyone here who is new, welcome!

This month, we just launched a new version of Unity AI, starting with Assistant. Participants of the new beta are able to test its greatly improved agentic capabilities, as well as upgrades to the generated asset types possible

Sign up here to try the latest version of Unity AI. Please note that the minimum version required is 6.3 for these features. Please ensure all fields are filled, as this information will be used to ensure your account is set up correctly.

If you’ve been a part of the beta so far, we sincerely thank you for your time and effort in helping us make Unity AI better. Points allocated to the legacy version of Unity AI will no longer refresh once the new beta goes live. Points will only be available using the new version of Unity AI, so we invite you to join here.

Thanks again for your continued support, and we’re excited to build the next phase with you.

Any issues, please feel free to post in thread or email us directly at ext-unity-ai-beta@unity3d.com

Unity AI Team

___________________________________________________________________________________________________________________

FAQ
What is included in the new Unity AI?

Assistant:

• Now has 2 modes, Ask and Agent. Ask is where you query for answers, and Agent routes your prompt to various dedicated individual agents to execute tasks in the Editor.
• Orchestration provides improved planning and reasoning. Basically, Assistant is able to break down complex tasks into multi-step resolutions as well as intelligently route your prompts to the right tools and agents. Meaning you can now prompt more complex questions, or have Assistant execute larger-scope actions for your project.
• Other features
  • Vision: Assistant can now use vision models to analyze images from your scene and project, giving agents relevant context for precise answers and improved comprehension.
  • Safer code: Assistant now recognizes project dependencies for safer code changes. With integrated Git support and a new visual diffing tool, you maintain full control over all AI-generated code before acceptance.
  • Focused context: Agents can index your project and classify assets to build a searchable knowledge graph. This ensures context sharing is intelligent, providing agents with only the necessary information to deliver focused answers even in complex projects.

Asset Generation

• Generators’ core capabilities, which already include sprites, textures, animations, audio, and materials are now expanded with powerful new generators
• UI Toolkit layouts: Generate UI Toolkit layouts from text prompts. Assistant produces ready-to-use UXML and USS that you can preview, edit, and utilize directly in your project.
• Cubemaps for skyboxes: The Skybox Generator allows you to create immersive 3D environments from a prompt or image. This tool transforms your input into high-quality cubemaps for skyboxes. You can generate custom, atmospheric environments.
• 3D meshes & textures: The 3D Model Generator empowers you to create production-ready 3D assets quickly from text prompt or reference image to a 3D model. Good for characters, props, or environment objects.

Why is the 2025 Beta ending?

The 2025 Beta has reached the end of its development cycle. We’ve reorganized our AI products into a more cohesive product, which requires the previous Beta packages to be retired in order to activate the new version.

Previous versions will no longer be active from Monday, Jan 12, 2026

How do I join the new Beta?

Apply via the sign-up form. Spaces are limited, so we recommend applying only if you can commit to regular interaction and feedback with the team.

Once your application is submitted, a member of staff will be in touch within 3-5 business days with an update on your status. We may contact active users from the previous 2025 beta for final feedback, or a short transition survey.

**Note: Unity version 6.3 or higher must be installed to use the new Beta packages.

What will happen to my existing work?

You can still access assets and conversations from the previous versions of Unity AI. However, as part of the transition, it will not be possible to generate anything new with points from the old versions.

What products will be available in the new Beta?

The AI Assistant and Generators will be available at launch. Additional products will be introduced in separate Beta groups.

What happens if I’m not accepted into the Beta?

If you’re not accepted initially, you’ll be placed on a waiting list. Additional openings may become available continuously as we ensure high engagement and participation with the packages during these trials.

Further questions can be submitted in the public Unity AI Discussions forum or our Discord.

Yes! We’ve made updates on the model side, but our public docs haven’t fully caught up yet, which is why it may look unchanged.

At a high level, we’re continuing to use the same “model family” in production for consistency, while rolling in newer versions and improvements behind the scenes (quality, reliability, and latency). We’re also validating changes carefully before we call them out publicly, so we don’t create confusion or promise something that isn’t broadly available.

We hear you. The “same model family in production” note is mainly about keeping behavior consistent while we’re still hardening quality, reliability, and evaluation across updates.

On adding other 3rd-party models, it’s definitely something we’re considering. In practice, different models tend to shine in different areas (for example, certain coding workflows vs. certain generation workflows), so we don’t assume a single model is “best” for everything.

For now we’re prioritizing stability and comparability, but we’re actively listening to feedback on where alternative models would add the most value. If you’re willing, sharing the specific coding and/or 3D workflows you’re benchmarking (and what “better” looks like for you) would help shape what we explore next.