I hit rate limits and “model is busy with other requests” frequently while just developing a highly concurrent agent app. Especially with the dated (e.g. -0613) or now -16k models.