Tok/s vs cost of self-hosting deepseek-ai/DeepSeek-V4.1-Flash on H100:8, H200:8.

Run it yourself

deepseek-v4.1-flash-h100@v2 · H100:8
$ iwant launch deepseek-v4.1-flash-h100
deepseek-v4.1-flash@v3 · H200:8
$ iwant launch deepseek-v4.1-flash

Results

Dashed lines estimate the same throughput at another cloud's price for the same GPUs.

setup$/hourpeak tok/s$/1M at peak
deepseek-v4.1-flash-h100@v2 · 8x H100 GCP europe-west2-b Spot$11.456,030$0.5275
deepseek-v4.1-flash@v3 · 8x H200 GCP asia-south1-b Spot$60.8911,328$1.4931
deepseek-v4.1-flash-h100@v2 · GCP europe-north1 spot (estimated)$8.706,030$0.4010
deepseek-v4.1-flash-h100@v2 · AWS us-east-1 on-demand (estimated)$55.046,030$2.5353
deepseek-v4.1-flash@v3 · AWS ap-northeast-2 spot (estimated)$12.1811,328$0.2988
deepseek-v4.1-flash@v3 · AWS us-east-1 on-demand (estimated)$63.3011,328$1.5521