Unverified50% confidenceBenchmarkExact time
DeepSeek V4.1 Flash reaches 2.52 million Tokens per second per GPU on NVIDIA VLR Rubin, 2.7 times that of GP300.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
11/14/2026
First Seen
Valid until: 11/28/2026
Sources
DeepSeek Opens Full App API Suite, Qwen4 Set for Nov 17 | AI Daily
bilibiliWill叔聊AI11/14/2026
Related Entities
Related Claims
Partially VerifiedOn a dual-Spark system, DeepSeek V4 Flash (FP8 quantization) peaks at approximately 53 tokens per second.79% similarUnverifiedDeepSeek V4.1 Flash has a composite score of 71.5, 11.4 points below GPT-6 Astra's 82.9, but its cost per task is only $0.27 and its output speed is about 235 tokens per second.76% similarVerifiedDeepSeek-V4.1-Flash's KV cache is compressed to 890 bytes per token, only one-quarter that of V4 Flash.76% similarUnverifiedDeepSeek-V4-Flash-Vision-Exp adopts a MoE architecture with approximately 305B total parameters, activating only approximately 13B per token, for an activation rate of approximately 4.6%.74% similarUnverifiedFlash 1017 can handle million-level context on a single A100-40GB GPU.74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1170493API
curl https://kongchang.com/api/v1/knowledge/claims/1170493MCP
get_claim(id=1170493)