← Leaderboard
nvidia/qwen3.6-27b

Qwen3 6 27B

vLLM 27B General
Throughput @ 4k
12.1t/s
@ 50k fill
8.1t/s
@ 100k fill
7.0t/s
Retention 100k÷4k
58%
Engine
vLLM
Parameters
27B
Released
2026-06-22

Headline numbers are perfbench-metrics (PBM): decode tok/s after a fixed context fill (4k / 50k / 100k), same agent-style workload as bench v2. Retention is how much of the 4k speed remains at 100k.

Fill curve

Throughput vs context fill — how the model holds up as the window grows.

14 7 0 12.1 4k 8.1 50k 7.0 100k tok/s

Golden profile

nvidia-qwen3-6-27b-eugr

Capabilities

explorervllm

Why we run it

HF Explorer download 2026-06-30

Bench notes

PBM 4k @ 12.1 tok/s — perfbench-metrics — profile=nvidia-qwen3-6-27b-eugr

Benchmarked 2026-07-10
SparkBench · GB10 · single node