Runway has announced Solaris, an "Interface World Model" that the company says can generate app-like experiences frame by frame without code. Alongside the product announcement, Runway released an internal benchmark in which Solaris reportedly outperformed Claude Opus 5 at following instructions.
What Runway presented
Solaris combines a vision model and a large language model (LLM): according to Runway, the LLM determines what should happen in a scene and the vision model renders the resulting frames. The company’s demo showcased a virtual clothing store where the system generates visual frames in sequence to support user interactions.
The benchmark and the criticism
Runway published its own benchmark claiming Solaris better follows instructions than Claude Opus 5. Critics have pointed out that Runway designed the benchmark and ran the test itself, and that Claude Opus 5 was not necessarily built for the same task Solaris targets. Observers argue that comparing Solaris to a model not optimized for the task can skew the outcome.
Previous pattern
Observers compare this announcement to a move last year involving Seedance, where Runway wrapped a partner model into a product presentation. The same pattern — packaging or benchmarking in a way that favors the company’s offering — is noted by some commentators.
Technological context
The underlying approach — pairing a vision model with an LLM that issues decisions and drives a renderer — is a feasible technical direction. A capable language model with tool access, connected to a rendering engine, could produce similar frame-by-frame interactions. The technology itself is real; the questions raised concern how results are framed and evaluated.
Why it matters
The story highlights two issues: the progress in multimodal systems that merge visual and language capabilities, and the importance of fair evaluation practices. When a company creates and owns the benchmark and compares its system to others not tailored for the task, conclusions about superiority should be treated with caution.
Conclusion
Solaris represents a real technical concept in the multimodal space and the demo illustrates its potential use cases. However, Runway’s benchmark and its declaration of victory warrant critical scrutiny because the evaluation was self-authored and compared Solaris to a model not explicitly designed for the same application.
