← Leaderboard
nvidia/qwen3.6-35b-a3b

Qwen3.6 250k NVFP4

vLLM MoE · 3B active / 35B total AgentsGeneralReasoning
Throughput
60.0t/s@ 250k
Engine
vLLM
Parameters
3B active / 35B total
Released
2026-05-27
Benchmarked
2026-06-27

Context ladder

Throughput at each benched context window (single measurement).

Context KV Throughput
@ 250k peak golden 60.0t/s

Golden profile

opencode-qwen36-250k

Capabilities

opencodeagentcodingnvfp4eugrgoldenvllm

Why we run it

Golden fleet target — auto-scaffolded from recipe opencode-qwen36-250k.

Bench notes

golden ?/? @ 60.0 tok/s — fill~50000 — bench-agent-v2 — tool_ok=True

Benchmarked 2026-06-27
SparkBench · GB10 · single node