WEBVTT

00:00:00.000 --> 00:00:08.000
A lower token quote is a starting point. The research question is the cost and value of useful work.

00:00:08.000 --> 00:00:20.000
$60.00 versus $12.83 per million output tokens. Launch price versus the BEP selected basket; different models and dates.

00:00:20.000 --> 00:00:31.000
About 79% lower quoted output price. Not evidence that the same workload became 79% cheaper.

00:00:31.000 --> 00:00:41.000
Fixed weights and model-priority rules. Not every model, not always the newest flagship.

00:00:41.000 --> 00:00:52.000
Include input/output, retries, tool calls, and human review. Compare at the same quality and latency target.

00:00:52.000 --> 00:01:02.000
Cheaper output tokens alone do not establish demand growth, realized margins, or investment returns.

00:01:02.000 --> 00:01:12.000
Next: repeatable task tests plus permissioned customer bills. Read the source notes at BEP Research.