TwelveLabs' Pegasus 1.5 arrives on Amazon Bedrock

TwelveLabs' Pegasus 1.5 will launch on Amazon Bedrock on March 15 in Brazil. Version 1.5 turns long videos into structured, timestamped metadata. Released on October 9, 2026.

Pegasus 1.5 processes videos without pre-indexing. It allows teams to analyze visual data more efficiently by converting unstructured content into organized information. The model's main feature is full video analysis, with segmentation, multimodal prompting with reference images and real-time processing. This means videos can be analyzed without the need for pre-indexing, which makes it easier to work with long content. The model can handle videos up to 2 hours long without losing analysis quality. Pegasus 1.5 is available via API on Amazon Bedrock, but only in the us-west-2 region.

This limits access for those located in that geographic area. The model is identified by the ID twelvelabs.pegasus-1-5-v1:0. The restricted availability can be an obstacle for teams that need to process videos in real time but are not in the us-west-2 region. TwelveLabs continues to expand its presence in the AI market, with Pegasus 1.5 integrated into the Amazon Bedrock ecosystem.

TwelveLabs' Pegasus 1.5 arrives on Amazon Bedrock — Image: aws.amazon.com

The maximum video length is a factor that changes the dynamics of use: it allows users to analyze longer content without losing accuracy. The model's availability is limited to the us-west-2 region, which can make access difficult for teams in other locations. These specifications define how Pegasus 1.5 differs from other video language models.

Previously, the model was already able to process videos, but version 1.5 brings significant improvements, such as full analysis with segmentation, multimodal prompting with reference images and real-time processing. In the market, rivals such as Meta's Llama Video and Alibaba's VideoLM also offer similar solutions, but Pegasus 1.5 handles long videos without preprocessing. The price in Brazil has not yet been disclosed.

The system identifies specific moments of a soccer match, such as substitutions and stoppage time. The output includes topics such as 'Substitutions', 'Stoppage time' and 'Penalty shootout preparation'. The model also recognizes final actions of the match, such as 'Final efforts and Final Whistle'. The model turns visual data into structured, timestamped metadata. This allows teams to work with real-time information without needing pre-indexing. The new feature can speed up processes in areas such as sports, entertainment and visual data analysis. Availability in Brazil has not yet been confirmed.