The Mechanism

• The Sync API moves transcription into a single request, removing the multi-step async overhead of uploading, polling for status updates, and manually retrieving results.0:34
• Developers can further reduce latency by using a pre-warm request to establish DNS, TCP, and TLS handshakes while the audio is still being recorded; this keeps connection setup off the transmission's critical path.1:59
• The product is intended for short, discrete tasks—like voice agents or live meeting notes—rather than batch jobs or continuous live streaming.2:53

Performance and Scope

• In a demo recording from Singapore routed to the EU via the global endpoint, the transcript returned in 1,900 ms, which the speaker presents as a success despite the long physical distance.3:44
• The company claims a P50 latency of approximately 134 ms when used within the US or EU, though this is a vendor-calculated metric rather than an independently verified result.4:21
• The sync workflow is supported by Universal 3.5 Pro, an AssemblyAI model the speaker asserts delivers the same accuracy as existing batch models, though this claim rests on vendor assertion rather than provided comparative methodology.