Source: aws.amazon.com/blogs/machine-learning/pegasus-1-5-from-twelvelabs-is-now-available-in-amazon-bedrock
Pegasus 1.5 is a video language model that analyzes long-form video and returns structured, timestamped metadata. Instead of answering a question about a clip, it reviews the full video and finds every segment that fits the user-defined criteria. Developers call it on Bedrock with the model ID twelvelabs.pegasus-1-5-v1:0 through InvokeModel or InvokeModelWithResponseStream.
Videos can be read directly from an Amazon S3 URI or sent as a base64 string of up to 25 MB, so no pre-indexing step is needed. A single request can cover videos up to two hours long, up from one hour in Pegasus 1.2. Segment definitions can combine a text description with up to 30 reference images, such as a product or logo.
Billing for Pegasus 1.5 runs through AWS Marketplace, so charges appear under TwelveLabs rather than Amazon Bedrock on the AWS bill and in Cost Explorer.
Why it matters if you build agents
Agents that process video can now receive structured segment data with start and end times instead of free-form text, which removes the need for a separate ingestion or indexing pipeline. Developers should plan around the limits, including 25 MB for inline base64 video and 2 GB for files read from S3. Teams tracking model spend should also note that Pegasus 1.5 charges appear under TwelveLabs in AWS billing.
Get stories like this every weekday
One short email with the day's agent tooling news.