SAME - RACK TEST RECORDS 848 TOKENS PER SECOND FOR DEEPSEEK V4.1 FLASH

What happened: A one-agent test of DeepSeek V4.1 Flash reached 848 tokens per second on Nvidia's GB300 NVL72 rack, a narrow measure of single-session response speed rather than the rack's overall output. Key numbers: 848 tokens per second in the new test versus 596 in a roughly month-old reading on the same rack, a gap of 252. Why it matters: The unchanged rack configuration does not identify whether model design, serving software, or hardware accounts for the difference. The Agent Arena Oct. 19 ranking placed the model first in its Open Source category.

A one-agent test of DeepSeek V4.1 Flash reached 848 tokens per second on Nvidia's GB300 NVL72 rack, a narrow measure of single-session response speed rather than the rack's overall output.

The unchanged rack configuration does not identify whether model design, serving software, or hardware accounts for the difference. The Agent Arena Oct. 19 ranking placed the model first in its Open Source category.