TwelveLabs Launches Pegasus 1.5, Turning Raw Video Into Structured, Queryable Data at Scale
Summary
TwelveLabs released Pegasus 1.5, a video reasoning model that extracts timestamped, structured metadata from videos up to two hours in a single API call. The model’s Time Based Metadata Extraction (TBM) accepts a user-defined JSON schema and returns production-grade, schema-compliant JSON with no ingestion pipeline, preprocessing, or indexing. Pegasus 1.5 supports multimodal prompting, on-the-fly processing from URLs or uploads, and demonstrates superior segmentation and structured-output reliability versus leading general-purpose models (13.1% segmentation improvement over Gemini 3.1 Pro in benchmarks). The model is available now via the TwelveLabs API and is being demonstrated at NAB Show 2026, with enterprise pricing and integration support through the company.