Kimi K2.5/K2.6 1T · GPU comparison

Kimi K2.5/K2.6 1T — B300 vs H200

Head-to-head AI inference benchmark comparison of B300 (NVIDIA Blackwell) and H200 (NVIDIA Hopper) on Kimi K2.5/K2.6 1T. Latency, throughput, and cost across LLM workloads. Use the chart controls below to switch sequences, precisions, and metrics — same interactions as the main inference chart.

B300 / H200 on Kimi K2.5/K2.6 1T at 51 tok/s/user: 464 / 310 tok/s/GPU, $1.41 / $1.24 per million tokens. H200 is 14% cheaper per token; B300 delivers 49% more tok/s/GPU.

Around the middle of the 34–105 tok/s/user interactivity band, at 69 tok/s/user on Kimi K2.5/K2.6 1T: B300 runs 330 tok/s/GPU at $1.96/M tokens, H200 runs 229 at $1.70/M. H200 is 15% cheaper per token; B300 delivers 44% more tok/s/GPU.

Setting 88 tok/s/user as the target on Kimi K2.5/K2.6 1T, B300 produces 233 tok/s/GPU ($2.81 per million tokens) and H200 produces 158 ($2.49). H200 is 13% cheaper per token; B300 delivers 47% more tok/s/GPU. (Numbers reflect the default 1k/1k · int4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)

View performance-per-dollar view →

Interpolated from real benchmark data. Edit target interactivity values below to compare at different operating points.
Metric
Interactivity (tok/s/user)
Interactivity (tok/s/user)
Interactivity (tok/s/user)
Throughput (tok/s/gpu)
B300:464.1H200:310.5
B300:330.2H200:229.0
B300:232.6H200:158.1
Cost ($/M tok)
B300:$1.412H200:$1.238
B300:$1.959H200:$1.704
B300:$2.809H200:$2.493
tok/s/MW
B300:213867H200:179479
B300:152176H200:132358
B300:107202H200:91376
Concurrency
B300:~32H200:~25
B300:~15H200:~13
B300:~5H200:~7

Inference Performance

Inference performance metrics across different models, hardware configurations, and serving parameters.