Runway’s Gen-4.5 Surpasses OpenAI and Google in Text-to-Video AI Race

New York-based AI company Runway has launched Gen-4.5, its most advanced text-to-video generation model to date, claiming top spot globally in video AI. According to the company’s announcement on December 4, 2025, Gen-4.5 is now rated higher than OpenAI’s Sora 2 and Google’s Veo 3, setting new standards in visual realism, physics accuracy, and creative control. Runway, which first made waves with its Gen-1 model in 2023, continues to push the boundaries of video generation with this latest iteration.

What Makes Gen-4.5 Stand Out?

The Gen-4.5 model is capable of creating videos with unmatched physical realism and cinematic quality, offering users granular control over motion, detail, and style.

It supports a broad spectrum of aesthetics — from photorealistic animation to Hollywood-style cinematic visuals — while maintaining high temporal consistency (i.e., smooth movement across video frames).

In benchmark tests, Gen-4.5 scored 1247 points in the Elo benchmark, edging out,

• Google’s Veo 3 (1226)
• OpenAI’s Sora 2 Pro (1206)

It was developed in collaboration with Nvidia and trained using the latest Blackwell and Hopper GPUs, allowing for efficient training and fast inference (video rendering).

Key Features of Gen-4.5

• Temporal Consistency: Maintains realistic frame-by-frame continuity in object motion, liquid flow, hair strand movement, and material dynamics.
• Controllable Action Generation: Users can input specific prompts for movements, scenes, or styles, allowing for directorial control over generated content.
• Photorealism with Speed: Despite generating highly detailed videos, Gen-4.5 doesn’t compromise on speed — it retains the performance efficiency of its predecessor, Gen-4.
• Creative Flexibility: Ideal for creators, designers, and filmmakers, the model supports a variety of applications from short films to design mockups, all from text prompts.

Limitations Still Persist

• Causality and Object Permanence: Gen-4.5 struggles with maintaining logical sequence of events in dynamic scenes, such as cause-effect relationships in actions.
• Success Bias: It may sometimes render idealized or incorrect outcomes — for instance, showing an arrow hitting the target despite being poorly aimed.
• Runway acknowledged these issues, stating they are part of the broader challenge of building reliable “world models” in AI. Continued iteration is expected to address these gaps.

Accessibility and Use

• Gen-4.5 is available across all subscription plans, making it more accessible to individual creators, studios, and enterprises.
• The model is designed to function seamlessly in commercial, educational, and creative environments where video generation from text can save time and expand visual storytelling.

Key Takeaways

• Runway’s Gen-4.5 outperforms Google’s Veo 3 and OpenAI’s Sora 2 Pro in text-to-video benchmarks.
• Elo benchmark score: Gen-4.5 (1247), Veo 3 (1226), Sora 2 Pro (1206).
• Offers unprecedented realism, temporal consistency, and style flexibility.
• Built with Nvidia GPUs (Blackwell and Hopper), enabling high performance and speed.
• Limitations include causality errors and success bias.
• Available across all subscription plans, accessible to a wide range of creators.