TwelveLabs, a leader in video AI, has announced the release of Pegasus 1.2, a cutting-edge multimodal foundation model that sets new industry standards in long-video understanding. Capable of processing videos up to one hour long with best-in-class accuracy, low latency, and cost efficiency, Pegasus 1.2 is designed for production-grade video processing pipelines that demand precision and scalability.

Breakthrough Capabilities of Pegasus 1.2

1. Unmatched Long-Video Understanding

• Processes videos up to 60 minutes long while maintaining high accuracy.
• Reduces computational costs with intelligent embeddings storage that caches videos for faster, low-cost repeated queries.

2. Advanced AI for Text Generation

• Converts video content into summaries, highlights, titles, and detailed reports with AI-powered language generation.
• Leverages both visual and audio analysis for deeper contextual understanding.

3. Optimized Computational Efficiency

• Uses advanced vision encoding and token reduction methods to balance computational efficiency and precision.
• Seamlessly scales from short video clips to full-length video analysis, making it ideal for real-world business applications.

Industry Impact: Elevating Video Intelligence

"Video understanding represents one of the most complex challenges in artificial intelligence," said Aiden Lee, CTO of TwelveLabs. "Pegasus 1.2 introduces a novel approach to spatio-temporal comprehension, addressing fundamental limitations in existing video language models."

By integrating with TwelveLabs' Marengo model, Pegasus 1.2 delivers human-like video intelligence, transforming how businesses extract insights from video content.

Why Pegasus 1.2 Matters for Businesses

Efficient Content Analysis – Automates video summarization and reporting for media, marketing, and research.
Scalable Performance – Adapts seamlessly to varying video lengths and complexities.
Cost-Effective AI Processing – Optimized architecture challenges the need for massive computational resources.

Pegasus 1.2 represents a milestone in multimodal AI, proving that superior performance can be achieved without exponentially increasing model sizes.