Runway Gen-4.5

What it is

Runway Gen-4.5, the flagship model from the American lab Runway, is best understood as a control surface rather than a raw quality play. It launched in late 2025 at 1247 Elo and was briefly first on Video Arena, with image-to-video support following in January 2026. Since then Runway Gen-4.5 has dropped out of the Artificial Analysis top ten as the Chinese models advanced, and any honest assessment has to start there: on the leaderboards alone, it is no longer the leader. The reason to keep paying attention lies elsewhere, in what Runway wraps around the model.

The control surface

That wrapper is the point. Runway sells a creative environment rather than a bare generation endpoint, and Runway Gen-4.5 sits at the centre of it. The toolkit includes keyframes for pinning down specific moments, a motion brush for directing movement by hand, video-to-video for reworking existing footage, camera controls for framing and moves, and storyboarding for planning a sequence before committing to a render. For a director or an editor who wants to shape a shot rather than roll the dice on a prompt, that degree of control is the whole proposition, and it is what the “control surface” label is meant to capture.

New audio and editing

The feature set grew meaningfully at the end of 2025. Alongside the GWM-1 announcement in December 2025, Runway added native audio generation, audio editing and multi-shot video editing, closing some of the gap with rivals that had shipped sound earlier and giving editors a way to work across several shots inside one environment. Audio editing in particular is a professional’s feature: being able to adjust a soundtrack in place, rather than exporting to another tool, is the kind of workflow detail that saves real time on a real project.

How pricing works

Pricing follows the software logic rather than the API one. Runway runs credit-based subscriptions from roughly $12 to $95 per month, so a user pays for access to the environment and works within a credit budget, instead of being metered by the second on every experiment. For a creator who iterates heavily, tries many versions and expects to throw most of them away, that model can be far kinder than per-second billing, where every abandoned take shows up on the invoice. This is the clearest reason to choose Runway Gen-4.5 over the per-second API vendors: predictable cost for heavy, exploratory work. For teams that measure their progress in iterations rather than in finished renders alone, that predictability is often worth more than a marginally higher score on a benchmark board, because it lets them keep experimenting without watching a meter tick over on every failed attempt.

Trade-offs and outlook

The trade-off is equally clear. A team that simply wants the highest possible output quality, generated at scale and pushed through an API into an automated pipeline, will find better raw scores among the Chinese leaders and a per-second model that suits batch production. Runway Gen-4.5 is not the tool for that shape of work. Its strengths lie in hands-on creative direction, in the user who cares more about steering the result than about topping a benchmark, and who values a single environment where planning, generation, editing and audio all live together.

Where this leaves Runway is an interesting question for the wider field. The company helped define generated video in the first place, and it has now been overtaken on pure quality by labs with larger data advantages. Its response has been to compete on craft rather than on raw scores, betting that professional users will pay for control and workflow even when a cheaper model can match the output. Whether that bet holds as the leaders keep improving is the thing to watch, and the GWM-1 line hints that Runway is not standing still. For how Runway Gen-4.5 compares with the current leaderboard, see our AI video models hub, part of the broader AI coverage.