Speed test matrix

command-a-plus-05-2026-fp8 on H200 NVL 141GB ×2

vllm · 0.26.1rc1.dev638+g52be12cfa · FP85 contexts (8K–123K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 123K, 128K

by @keennaylmx dev importMeasured Sep 2, 2026, 23:50 UTC

Hardware

Type: Discrete GPU
GPU: H200 NVL
GPU count: 2
VRAM: 282 GB (2 × 141 GB)
OS: linux

Engine

Engine: vllm
Version: 0.26.1rc1.dev638+g52be12cfa
Quantization: FP8
Backend: cuda
Repository: github.com/vllm-project/vllm
Build: cuda-13.2
Commit: 52be12cfac0c

Run

Served model: command-a-plus
Revision: main
Benchmark tool: llm-inference-bench v0.4.29
Decode test: Fixed duration
Prefill mode: Measured during decode warm-up
Time per measurement: 30 s
Max output: 4,096 tokens

Launch command

"$HOME/env_h200-cohere-vllm/bin/vllm" serve CohereLabs/command-a-plus-05-2026-fp8 --served-model-name command-a-plus --tensor-parallel-size 2 --reasoning-parser cohere_command4 --enable-auto-tool-choice --tool-call-parser cohere_command4 --max-model-len 131072 --gpu-memory-utilization 0.862577 --host 0.0.0.0 --port 8000 --api-key <redacted>

Prefill

prompt tok/s by context length

8K context19.6K419 ms TTFT

16K context18.4K874 ms TTFT

32K context17.1K1,873 ms TTFT

64K context15.3K4,172 ms TTFT

123K context12.4K9,837 ms TTFT

128K context12.3K10,362 ms TTFT

Sustained decode

total tok/s · rows are context length, columns are simultaneous requests

Context	1	2	4	8	16	32	64	128
8K	124.768 ms TTFT	211.086 ms TTFT	345.5108 ms TTFT	542.8130 ms TTFT	840.9243 ms TTFT	1,344.3391 ms TTFT	Over capacity	Over capacity
16K	123.997 ms TTFT	210.998 ms TTFT	339.4172 ms TTFT	533.2219 ms TTFT	833.6272 ms TTFT	Over capacity	Over capacity	Over capacity
32K	122.8141 ms TTFT	208.6133 ms TTFT	332.7205 ms TTFT	525.7307 ms TTFT	811.0431 ms TTFT	Over capacity	Over capacity	Over capacity
64K	121.2188 ms TTFT	202.3253 ms TTFT	326.6337 ms TTFT	503.0601 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
123K	117.8311 ms TTFT	196.9435 ms TTFT	309.7575 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity

Over capacity: not run because this many requests at this context would exceed the server's KV cache.

TTFT is measured after the benchmark warms the server's prompt cache, so it mostly excludes prefill.

All measurements

Latencies are medians. Total tok/s is output across all simultaneous requests.

Phase	Context	Requests at once	Status	Total tok/s	Per request tok/s	TTFT	Time between tokens	Average active requests	Completed requests	Output tokens	Measured for
Prefill	8K	1	Valid	19.6K	—	419 ms	—	—	1	—	—
Prefill	16K	1	Valid	18.4K	—	874 ms	—	—	1	—	—
Prefill	32K	1	Valid	17.1K	—	1,873 ms	—	—	1	—	—
Prefill	64K	1	Valid	15.3K	—	4,172 ms	—	—	1	—	—
Prefill	123K	1	Valid	12.4K	—	9,837 ms	—	—	1	—	—
Prefill	128K	1	Valid	12.3K	—	10,362 ms	—	—	1	—	—
Sustained decode	8K	1	Valid	124.7	127.0	68 ms	7.9 ms	1.0	1	3,740	30.0 s
Sustained decode	8K	2	Valid	211.0	105.8	86 ms	9.4 ms	2.0	0	6,330	30.0 s
Sustained decode	8K	4	Valid	345.5	86.7	108 ms	11.5 ms	4.0	0	10,364	30.0 s
Sustained decode	8K	8	Valid	542.8	68.8	130 ms	14.5 ms	8.0	0	16,279	30.0 s
Sustained decode	8K	16	Valid	840.9	53.6	243 ms	18.7 ms	16.0	0	25,216	30.0 s
Sustained decode	8K	32	Valid	1,344.3	42.9	391 ms	23.3 ms	32.0	0	40,319	30.0 s
Sustained decode	8K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	8K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	123.9	128.3	97 ms	7.8 ms	1.0	1	3,719	30.0 s
Sustained decode	16K	2	Valid	210.9	105.7	98 ms	9.5 ms	2.0	0	6,329	30.0 s
Sustained decode	16K	4	Valid	339.4	85.4	172 ms	11.7 ms	4.0	0	10,187	30.0 s
Sustained decode	16K	8	Valid	533.2	67.7	219 ms	14.8 ms	8.0	0	15,992	30.0 s
Sustained decode	16K	16	Valid	833.6	53.0	272 ms	18.9 ms	16.0	0	25,008	30.0 s
Sustained decode	16K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	122.8	164.5	141 ms	6.5 ms	1.0	1	3,685	30.0 s
Sustained decode	32K	2	Valid	208.6	104.5	133 ms	9.6 ms	2.0	0	6,258	30.0 s
Sustained decode	32K	4	Valid	332.7	83.9	205 ms	11.9 ms	4.0	0	9,980	30.0 s
Sustained decode	32K	8	Valid	525.7	66.5	307 ms	15.0 ms	8.0	0	15,768	30.0 s
Sustained decode	32K	16	Valid	811.0	51.4	431 ms	19.5 ms	16.0	0	24,319	30.0 s
Sustained decode	32K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	1	Valid	121.2	121.5	188 ms	8.2 ms	1.0	0	3,637	30.0 s
Sustained decode	64K	2	Valid	202.3	101.6	253 ms	9.8 ms	2.0	0	6,068	30.0 s
Sustained decode	64K	4	Valid	326.6	82.1	337 ms	12.2 ms	4.0	0	9,795	30.0 s
Sustained decode	64K	8	Valid	503.0	63.5	601 ms	15.7 ms	8.0	0	15,088	30.0 s
Sustained decode	64K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	123K	1	Valid	117.8	118.1	311 ms	8.5 ms	1.0	0	3,534	30.0 s
Sustained decode	123K	2	Valid	196.9	98.9	435 ms	10.1 ms	2.0	0	5,910	30.0 s
Sustained decode	123K	4	Valid	309.7	78.1	575 ms	12.8 ms	4.0	0	9,292	30.0 s
Sustained decode	123K	8	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	123K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	123K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	123K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	123K	128	Over capacity	—	—	—	—	—	—	—	—

Over capacity: not run because this many requests at this context would exceed the server's KV cache.