DeepSeek V4 Flash makes no sense on Foundry
retained text
Everyone uses agents, an agent makes anywhere between 1-10 requests to finish doing the task it needs to do. DeepSeek V4 Flash is an extremely fast and cheap model for agentic use. Tell me why the hell they decided to set a default max RPM limit of 20 on DeepSeek V4 Flash deployments. I can't even last more than 2 requests before being throttled. For a model BUILT for agentic uses, how does it make sense to set quota on requests instead of tokens???
document fetched 2026-09-07 00:34 · canonical https://reddit.com/r/AZURE/comments/1w9075d/deepseek_v4_flash_makes_no_sense_on_foundry