Runway Gen-4.5 overtakes Google Veo 3

Runway is back to the top with its latest Gen-4.5 AI Video model, surpassing Google's Veo 3.

Runway has officially shaken up the video AI landscape with Gen 4.5, a model developed by a team of around 100 researchers that has unexpectedly overtaken both Google and OpenAI on the industry’s leading blind video benchmark. The system uses a new Autoregressive-to-Diffusion (A2D) architecture that blends the logical sequencing strengths of autoregressive models with the visual fidelity of diffusion, giving Runway a major leap in realism, object permanence, and motion physics.

This hybrid approach allows the model to understand scenes through a vision-language backbone while rendering frames in parallel, avoiding the typical flickering and distortion seen in competing models. As a result, Gen 4.5 maintains consistent lighting, fluid dynamics, and character movement, addressing long-standing weaknesses in generative video. Benchmarks back this up: on the Video Arena leaderboard, Gen 4.5 leads with an Elo score of 1247, ahead of Google’s Veo 3, Kuaishou’s Kling 2.5, and even OpenAI’s Sora 2, which has slipped to seventh place.

Introducing our new frontier video model, Runway Gen-4.5. Previously known as Whisper Thunder (aka) David.

Gen-4.5 is state-of-the-art and sets a new standard for video generation motion quality, prompt adherence and visual fidelity.

Learn more below. pic.twitter.com/2J945S4bQt

— Runway (@runwayml) December 1, 2025

Gen-4.5 achieves an unprecedented level of physical accuracy and visual precision. Objects move with realistic weight and momentum. Surfaces behave the way they would in the real world. Laws of physics can be both observed or ignored, depending on your vision.

Despite the big… pic.twitter.com/AsbLQ0ZSDH

— Runway (@runwayml) December 1, 2025

The model isn’t perfect — notably, it lacks built-in audio generation, leaving creators to rely on external tools for dialogue, sound effects, or ambience. That gives Google’s Veo an advantage in convenience, even if Runway currently dominates on visual quality and prompt accuracy. Still, the platform includes useful style presets such as Photorealistic, Cinematic, and Slice of Life, and it demonstrates surprising improvements in physics and object permanence that set a new standard for the field.

Looking back at how things were just two years ago, when Runway Gen-1 was launched, it makes us wonder how things will be two years from now.

Two years ago, we introduced Gen-1, the first publicly available video generation model. It enabled an entirely new form of creative expression and created a new product category.

Over the past two years, we've led the industry in making video models more powerful and… pic.twitter.com/eLJfFtvQUg

— Runway (@runwayml) December 1, 2025

Runway Gen 4.5 pricing reflects the high cost of running on NVIDIA H100 and Blackwell GPUs. Plans start with a free tier offering 125 credits (enough for short tests), while paid tiers scale up to the $76/month Unlimited plan, which is positioned as the best option for heavy creators thanks to unlimited, throttled rendering. With its strong technical leap and clear win over Google Veo 3 in blind evaluations, Gen 4.5 solidifies Runway as a serious challenger to the biggest players in AI video.