Pegasus 1.5

Pegasus 1.5 is a specialized video-to-text API model rather than a general-purpose frontier model. Official documentation supports prompt-based video analysis, summaries, segmentation, and time-coded JSON metadata for videos up to two hours without pre-indexing. That gives it stronger video I/O fit than Bria 3.2’s 78 multimodal score, but the supplied evidence does not establish broad multimodal, language, or general technical capability comparable to Gemini 3.5 Flash or Claude Opus 4.8. Coding and agentic scores are deliberately low: DeepSWE and LiveCodeBench were checked and list no Pegasus/TwelveLabs result; no SWE-bench result was found. The same absence limits general reasoning claims. Official API documentation and the model’s structured time-based outputs support a usable developer experience, although there is no supplied latency, uptime, throughput, or independent deployment evidence to support the speed and availability levels assigned to GPT-5.6 Luna or Gemini 3.5 Flash. Pricing was not supplied in official evidence, making price clarity materially weaker than published API anchors. Cost effectiveness is therefore provisional rather than a claim of low value. Public signal is moderate: TwelveLabs has official product, documentation, and GitHub presence, but no supplied LMArena/Arena-Hard preference result, independent benchmark performance, or adoption metric. Scores remain below broadly evidenced generalist leaders while recognizing a credible, documented video-analysis specialization.