Runway Gen 4.5 Takes Top Spot in AI Video Generation, Beating Google and OpenAI

AI startup Runway has launched Gen 4.5, a groundbreaking video generation model that has immediately claimed the number one position on the prestigious Video Arena leaderboard, outperforming offerings from tech giants Google and OpenAI.

The model, internally codenamed "David" in reference to the biblical underdog story, represents a remarkable achievement for the 100-person company competing against trillion-dollar corporations. Runway CEO Cristóbal Valenzuela emphasized that the breakthrough demonstrates how focused innovation can overcome resource disadvantages.

We managed to out-compete trillion-dollar companies with a team of 100 people. You can get to frontiers just by being extremely focused and diligent, Valenzuela explained in an interview.

Gen 4.5 achieved 1,247 Elo points on the Artificial Analysis Text-to-Video benchmark, surpassing Google's Veo 3 at 1,226 points and ranking ahead of all other available AI video models. The Video Arena leaderboard uses blind testing where users compare anonymously labeled video clips side by side, eliminating brand bias and marketing influence.

The independent benchmark has become the gold standard for evaluating video generation quality, making Runway's top placement particularly significant for industry credibility and customer acquisition.

Gen 4.5 demonstrates exceptional capabilities across multiple dimensions of video generation:

Physics Understanding: The model accurately simulates real-world physics, from objects falling and shattering convincingly to rain droplets sliding down windows with proper motion and refraction.

Human Motion: Characters move with natural weight distribution and realistic biomechanics, eliminating the floating or disconnected appearance that plagued earlier models.

Camera Work: The system understands cinematography principles, executing complex camera movements with proper parallax, motion blur, and tracking that would satisfy professional cinematographers.

Prompt Adherence: Gen 4.5 excels at translating detailed written instructions into accurate visual sequences, allowing creators to specify intricate camera choreography, scene compositions, precise timing, and atmospheric changes within single prompts.

The model generates high-definition video with cinematic quality and unprecedented visual fidelity. Early adopters report that Gen 4.5 produces output that closely matches professional production standards, potentially bridging the gap between imagination and finished footage.

Runway's customer base already includes major brands like New Balance and Under Armour, media organizations, advertising agencies, and filmmakers. The company's tools have been deployed in production workflows for programs including The Late Show With Stephen Colbert and in Oscar-nominated projects.

Gen 4.5 was developed entirely on NVIDIA GPUs, including the Hopper and Blackwell series, with extensive collaboration to optimize video diffusion models for both training efficiency and inference speed. The approach balances performance with quality, delivering what Runway describes as an industry-first optimization.

The model represents significant advances in pre-training data efficiency and post-training techniques, building on seven years of focused research since the company's founding in 2018.

Despite the substantial capabilities leap, Runway acknowledges persistent challenges common to current video generation technology:

Causal Reasoning: Effects sometimes precede causes, such as doors opening before handles are pressed.

Object Permanence: Objects may disappear or appear unexpectedly across frames when temporarily obscured.

Success Bias: Actions tend to succeed disproportionately, like poorly aimed kicks still scoring goals.

These limitations are particularly important for Runway's work on world models, which require accurate representation of action outcomes in environments.

Founded in 2018 by Cristóbal Valenzuela, Alejandro Matamala, and Anastasis Germanidis at NYU's Tisch School of the Arts, Runway has evolved from a model directory into a comprehensive creative suite. The company reported earning 80 million dollars in annualized recurring revenue in December 2024, with projections to reach 300 million by the end of 2025.

The startup's valuation has grown to approximately 3.55 billion dollars, backed by investors including General Atlantic, Baillie Gifford, NVIDIA, and Salesforce Ventures. With roughly 100,000 customers, Runway has demonstrated commercial viability while competing against much larger players.

Gen 4.5 is being rolled out gradually and will be available to all Runway customers by the end of this week. The model will be accessible through Runway's platform, its application programming interface, and through select partners.

Valenzuela indicated that Gen 4.5 represents the first of several major releases the company has planned, suggesting continued innovation momentum.

The launch signals what some are calling the efficiency era of artificial intelligence, where breakthrough innovations don't necessarily require massive resource investments. Runway's achievement demonstrates that specialized focus and technical excellence can overcome the advantages of virtually unlimited budgets and large research teams.

For the creative industry, Gen 4.5 may represent a turning point where generative video tools transition from experimental curiosities to professional production tools that creators genuinely prefer over traditional methods.

Valenzuela expressed broader aspirations for the technology landscape, stating the company is excited to ensure that AI is not monopolized by two or three companies.