Use-case preset
Meeting transcript summarization cost calculator
Zoom/Meet transcript to action items and summary in batch.
Meeting transcript summarization ingests a raw Zoom or Google Meet transcript — 10,000 to 28,000 tokens of speaker-labeled dialogue — and emits a structured summary with action items, decisions, and owners typically under 800 tokens. The 95/5 input/output split and 32k context are driven by transcript length; output is intentionally constrained to a fixed template.
Batch scheduling removes any latency pressure, enabling best-effort throughput pricing. cachedPromptPercent stays low at 10 because each transcript is unique and dominates the prompt. The main cost lever is chunking: splitting a two-hour transcript into 30-minute segments and running four parallel jobs can reduce peak context per call without degrading summary quality. Models with strong instruction following at 32k+ context outperform raw quality scores here; 70B models are usually sufficient.