What changed
Amazon Bedrock made TwelveLabs Pegasus 1.5 available for on-demand video understanding. The model can process up to two hours without pre-indexing, return timestamped segments and structured JSON, and use reference images alongside text prompts.
Why it matters
The release turns long-form video search and cataloging into a managed inference call instead of a separate ingestion pipeline. It is useful for archives, sports and brand-monitoring workflows, but concentrates quality, access and cost decisions in one hosted model and region path.
What to do
Enable Pegasus 1.5 in a test Bedrock account, define one segment schema, replay a licensed representative video and compare timestamps, missed events, false matches, latency and metering with a human-reviewed reference set before automating downstream actions.
Caveat
Capability and data-boundary statements are AWS and TwelveLabs claims, not an independent evaluation. Initial direct model access is documented in us-west-2 with US and Global cross-Region profiles, and quality, price, quotas, content rights and regional processing still require tenant-specific checks.