Runway Solaris

Overview

Runway Solaris is the first model in Runway's Interface World Model family. Announced on August 31, 2026, it generates an interactive interface as a continuous stream of visual frames instead of producing HTML, components, or another coded representation. A starting image defines the initial state, and the model responds as a person clicks, drags, types, or gives natural-language instructions.

Solaris is not a replacement for Runway Gen-4.5, the generally available video model on paid Runway plans. It is an early-access research model built on Gen-4.5 and informed by Runway GWM-1. Its current value is in generated prototypes, interactive visual experiences, and variable environments for training computer-use agents.

What's New

The Interface Is the Model Output

Solaris removes the normal handoff from a visual design to a coded implementation. The world model renders the complete interface directly, preserving visual detail that can be lost when a screenshot or mockup is translated into layouts, components, and rules. There are no predefined screens or templates behind the generated experience.

This approach is most relevant when the interface itself should remain visual and adaptive. Runway demonstrates product showrooms, interactive tutorials, scientific demonstrations, and scenes where objects can be manipulated directly rather than controlled through fixed menus.

Real-Time Action Conditioning

Clicks, drags, typed input, and prior frames condition what Solaris renders next. The model learned from interactions that had already occurred, allowing it to connect an action with a plausible visual response without requiring each behavior to be programmed in advance. Runway says the system was designed around interactive response speed, session coherence, and visual quality that holds at 720p.

Solaris converts the Gen-4.5 diffusion process into an autoregressive real-time engine. Runway describes a three-stage process: generating frames sequentially, distilling many denoising steps into a few steps, and training the faster model on its own outputs to improve stability during continued interaction.

Separate Reasoning From Rendering

An accompanying language model decides how the interface should evolve. It interprets the request, determines whether an action changes the current scene or moves to a new state, and produces instructions for Solaris. Solaris then renders the visual response frame by frame.

This division gives the system a different role from an LLM that generates frontend code. The language model determines behavior, but the visible result remains a generated visual world rather than a conventional application tree.

Open-Ended Interaction

Traditional interfaces expose only the behaviors that developers implemented. Solaris can assign behavior through natural-language instructions, allowing the same visual scene to support different interactions. An object can become a tool, a tutorial can react to an unexpected action, or a storefront can reorganize around a user's request without switching to a predefined page.

This flexibility is also relevant to agent training. Instead of repeatedly training a computer-use agent against one fixed layout, teams could expose it to interfaces that change while preserving the task and underlying visual context.

Performance Benchmarks

Runway compared Solaris with interfaces generated as code by Claude Opus 5. The study used 250 participants, 30 interaction examples, and nearly 7,500 pairwise judgments.

These are Runway's own launch results, not an independent benchmark. They evaluate preference for visual interaction behavior, not accessibility, application correctness, latency under production traffic, or completion of normal software tasks.

Compared With GWM-1 and Gen-4.5

Solaris is a parallel model in Runway's research portfolio rather than the next numbered Gen model.

Teams should not treat Solaris as an upgrade from Gen-4.5 for normal video production. Gen-4.5 remains the practical choice for generating deliverable clips, while Solaris targets interfaces that must remain interactive after the first frame.

Availability & Access

Solaris is not generally available in the Runway web app or public model API. Runway is working with selected partners and provides an application form on the official announcement page. Applicants submit a business email, company information, and details about the intended use case.

Runway has not announced a public model identifier, API contract, subscription entitlement, credit rate, or general-availability date. Teams evaluating Solaris should therefore treat access, capacity, integration requirements, and production pricing as unresolved until Runway provides partner-specific terms.

Who Should Apply / Who Should Wait

Apply for early access if you are:

• Prototyping visual shopping, education, simulation, or product experiences where users manipulate a generated scene directly.
• Training computer-use agents and need layouts that vary instead of repeating one coded interface.
• Researching real-time world models, visual interaction, or alternatives to code-generated prototypes.
• Able to test experimental output without relying on stable, legible real-time text, long-session coherence, or mature accessibility integration.

Wait for broader availability if you need:

• A production website or app that requires stable real-time text, long-session coherence, or mature accessibility and software-stack integration.
• Legible text, keyboard and screen-reader compatibility, or integration with accessibility APIs.
• Published API specifications, service-level commitments, regional availability, or predictable pricing.
• Long sessions where visual identity, layout, and semantic state must remain exact.

Known Limitations

• Text rendering: Runway says stable, legible real-time text remains an open problem. Text-heavy experiences may need a hybrid image or coded interface.
• Trust and grounding: Solaris is primarily anchored by its starting state and supplied references. Plausible visual output is not proof that instructional or commercial information is correct.
• Long-session coherence: Visual and semantic consistency can degrade over extended, open-ended interactions.
• Accessibility and integration: Generated visual interfaces still need a way to work with screen readers, accessibility APIs, and the rest of a conventional software stack.
• Early-access uncertainty: Public pricing, API details, availability, and production operating limits have not been announced.

FAQ

Sources

• Introducing Solaris - official announcement, architecture, evaluation, limitations, and early-access form.
• Introducing Runway GWM-1 - official world-model lineage and GWM-1 capabilities.
• Introducing Runway Gen-4.5 - official base-model and availability context.