ModelCards

Pegasus 1.5

Video modelPrevious generationCard summarized

Pegasus 1.5 is a generative model for video-to-text generation.

Source: Building Pegasus 1.5: From Clip-Based QA to Time-Based Metadata (technical launch post), Pegasus 1.5 docs page and release notes (no model card published)

Use it via API

$1.75 / hour
Code exampleCompare

Where to call it

  • TwelveLabs APIpegasus1.5
  • Amazon Bedrocktwelvelabs.pegasus-1-5-v1:0

Supports

  • Streaming
  • Batch API
  • Image input
Full feature support

Good for

  • Segmenting editorial packages, scenes, sports plays, speaker changes or brand appearances into timestamped JSON
  • Summaries and detailed descriptions of long videos
  • Answering when specific events occur

Watch out for

Replaced by

Pegasus 1.6

Compare side by side

Overview

TwelveLabs · 02Pegasus, front

Pegasus 1.5

Pegasus 1.5 is a generative model for video-to-text generation.

Typical price
$1.75 / hour
Video input (analysis)
Max resolution
5184×2160
Max length
2 h analyzed per request
Modes
video-to-text, video Q&A
Released
20 Apr 2026

Pegasus 1.5 moved TwelveLabs' video-language model from answering questions about a clip (Pegasus 1.2) to partitioning a whole video by a user-defined schema and labeling each segment with structured, timestamped metadata. It also does prompt-based analysis with reference images and video clipping, over a 261,120-token context covering up to 2 hours of video. It was trained with reinforcement learning on verifiable rewards built from its own segmentation metrics, and Pegasus 1.6 (6 Oct 2026) now supersedes it. Technical launch post; release notes (20 Apr, 27 Apr, 6 May, 28 May 2026)

pegasus1.5
Typical price
$1.75 / hour
Max resolution
5184×2160 (input video)
Max length
2 h analyzed per request (source up to 4 h; 1 h on the sync endpoint)
Languages
English (full); Arabic, Chinese, French, German, Italian, Japanese, Korean, Portuguese, Russian, Spanish, Thai, Vietnamese (partial)
Weights
Closed
Released
20 Apr 2026
  • video-to-text
  • video Q&A
  • video summarization
  • video segmentation (timestamped JSON)
  • temporal grounding
  • image-guided prompts
  • structured JSON output