This section explains how Provisioned Throughput works for the Veo 3 and later models, including quota enforcement behavior and minimum GSU purchase increment.
For Veo 3 models, the quota enforcement period varies based on the number of GSUs you purchase for the model, and is subject to change. The quota enforcement periods for different GSU sizes are as follows:
1-9 GSUs: 2000 seconds
10-19 GSUs: 400 seconds
20-39 GSUs: 200 seconds
40-66 GSUs: 100 seconds
67 or more GSUs: 60 seconds
Note that this isn't connected to the request latency. The time to process your request isn't the same as the quota enforcement period.
For example, if you have a workload that requires generating a four-second video on the Veo 3 model and you purchase 1 GSU, you can generate that video within a few minutes. However, because the enforcement window for 1 GSU is 2000 seconds, you can't generate a video of the same size until the end of that period. Note that this is subject to Vertex AI clock time and doesn't depend on the time of your request. We recommend using the estimation tool on the Provisioned Throughput page to estimate the number of GSUs required for your workload. For more information about using the estimation tool, see Purchase a Provisioned Throughput order. These large enforcement windows ensure that your request is processed within a specific time. If you need to process outputs on Veo 3 more frequently, you must buy enough GSUs and check the corresponding enforcement window.